{ "data": { "a": { "slug": "muse-spark-open", "name": "Muse Spark Open", "vendor": "Meta", "vendorUrl": "https://dev.meta.ai/models/muse-spark-open", "kind": "model", "category": "inference", "summary": "Muse Spark Open is Meta's open-weight multimodal model for coding and agentic work, with 816B total and 42B active parameters and a 1M-token context, released on 15 October 2026 under the Muse Model Licence for self-hosting.", "url": "https://www.anchorterminal.com/tools/muse-spark-open", "markdownUrl": "https://www.anchorterminal.com/tools/muse-spark-open.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/muse-spark-open.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/muse-spark-open.json", "dns": "muse-spark-open.grade.anchorterminal.com", "repo": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "license": "Muse Model Licence (source-available weights, not an OSI licence), free of charge with an attribution notice, a separate agreement with Meta before Model as a Service by a licensee group above $20M gross revenue, and the Muse Spark Open Usage Policy", "transports": [ "http" ], "packages": [], "auth": "none", "authNotes": "Running the model needs no account or key. Getting Meta's weights does. Both Hugging Face repositories (BF16 and NVFP4) are gated with manual review and ask for full legal name, date of birth, country, affiliation and job title, and downloads then use a Hugging Face access token. Meta's Get the model page also gives a download from its own site after signing in and accepting the licence, one file at a time. The GGUF quantisations from ggml-org and Unsloth that Meta's docs point to are not gated. A local server has whatever authentication the owner adds, such as llama.cpp's `--api-key`.", "pricing": "free", "pricingNotes": "Free to download and run under the Muse Model Licence, with the owner's GPUs as the cost. A licensee whose group has more than $20M gross revenue in any 12 months must notify Meta and sign a separate agreement before offering it as Model as a Service. Hosted copies cost $1.25 input, $0.15 cached input and $4.25 output per million tokens on Fireworks, Baseten and Modal (checked 2026-10-15). On 4 November Hugging Face lists Baseten and Together as live inference providers, Baseten at $4.25 output. Meta Model API does not serve it.", "priceSummary": "from $1.25 / 1M in", "where": "local", "x402": { "level": "no", "evidence": "No x402, MPP or L402. Muse Spark Open is a model the owner runs, with no payment route from Meta, and Meta Model API does not serve it (checked 2026-10-15).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-10-15" }, "docsUrl": "https://dev.meta.ai/docs/muse-spark-open", "llmsTxt": "https://dev.meta.ai/llms.txt", "capabilities": [ "inference.llm", "inference.open-weights", "inference.local" ], "tags": [ "model", "open-weights", "self-hosted", "local", "free", "gated", "multimodal", "openai-compatible", "llms-txt", "gpu", "new" ], "lastRelease": "2026-10-15", "graded": true, "anchor": { "graded": true, "score": 53.4, "grade": "D", "agentReady": false, "rank": 845, "ranked": true, "rankOf": 1350, "categoryRank": 25, "methodology": "0.7", "run": "2026-10-01", "scores": { "ergonomics": 69, "maintenance": 60, "payments": 50, "reliability": 42, "schema": 61, "security": 44, "transparency": 53 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "medium", "date": "2026-11-04" }, "negative": 0, "verdict": "Muse Spark Open runs on the owner's hardware in seven runtimes, with OpenAI-compatible servers in vLLM and SGLang, standard tool calls, a 1M-token context and Meta's own serving docs. Serving the 537 GB NVFP4 checkpoint takes four B300 GPUs, Meta's download is gated by manual review, and the licence adds attribution and a revenue threshold for model resellers.", "bestFor": "A team with a four-GPU Blackwell node, or two to four Mac Studios, that wants a 1M-token multimodal coding and agent model on its own hardware with no prompts leaving it.", "strengths": [ "816B total and 42B active parameters with a 1,048,576-token context, taking text, image, video and audio input", "Day-one support in vLLM, SGLang, TensorRT-LLM, Transformers, llama.cpp, MLX and Unsloth, with commands for each in Meta's docs", "With the `muse` parsers, reasoning comes back in its own field and tool calls as a standard `tool_calls` array", "The licence grants use, fine-tuning, distillation and resale free of charge, with no user cap and no regional exclusion", "llms.txt and Markdown docs cover checkpoints, prompting, stop tokens, recommended settings and post-training" ], "weaknesses": [ "The NVFP4 checkpoint (about 537 GB) serves on four B300 GPUs and the BF16 one (about 1.65 TB) on eight", "Meta's Hugging Face repositories need manual approval with legal name and date of birth, which the Get the model page doesn't mention", "Products must display \"Built with Muse Spark Open\", and a company group with over $20M revenue needs Meta's agreement to offer it as Model as a Service", "The Usage Policy rules out military, nuclear, espionage, critical infrastructure and heavy machinery uses, and users under 18", "No changelog, deprecation policy or CI found, and none of the Hugging Face discussion replies counted on 4 November came from Meta" ], "agentNotes": [ "Start vLLM or SGLang with `--reasoning-parser muse --tool-call-parser muse` (vLLM also `--enable-auto-tool-choice`), or tool calls come back as raw text", "Set `reasoning_effort` to `low` or `medium` for short tasks. The default is `xhigh`, reasoning can't be turned off, and other values return an error", "Stop only on `\u003c|eot|\u003e` and end of text. Stopping on `\u003c|eom|\u003e` cuts off tool calls and final answers", "Stream long generations and leave `max_tokens` room for reasoning, or `content` can come back empty with `finish_reason` `length`", "Don't call Meta Model API for this model. Use a self-hosted server or a hosted provider's model string, such as `meta-models/muse-spark-open`" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "D", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 53.4 } ], "editorialScores": { "ergonomics": 69, "maintenance": 60, "payments": 50, "reliability": 42, "schema": 61, "security": 44, "transparency": 50 }, "provenanceScore": 55, "checked": { "categories": { "ergonomics": "2026-11-04", "maintenance": "2026-11-04", "payments": "2026-11-04", "reliability": "2026-11-04", "schema": "2026-11-04", "security": "2026-11-04", "transparency": "2026-11-04" }, "last": "2026-11-04", "lastMoved": "2026-10-30", "oldest": "2026-11-04", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "hf download meta-models/Muse-Spark-Open-816B-A42B-NVFP4 --local-dir ./muse-spark-open-nvfp4\n\nexport VLLM_USE_RUST_FRONTEND=1\nvllm serve meta-models/Muse-Spark-Open-816B-A42B-NVFP4 \\\n --tensor-parallel-size 4 \\\n --enable-expert-parallel \\\n --max-model-len 1048576 \\\n --reasoning-parser muse \\\n --tool-call-parser muse \\\n --enable-auto-tool-choice", "http": "curl http://localhost:8000/v1/chat/completions \\\n -H 'Content-Type: application/json' \\\n -d '{\n \"model\": \"meta-models/Muse-Spark-Open-816B-A42B-NVFP4\",\n \"messages\": [{\"role\": \"user\", \"content\": \"Refactor this module.\"}],\n \"reasoning_effort\": \"xhigh\"\n }'" }, "letme": { "capability": "https://letme.dev/inference.llm", "tool": "https://letme.dev/muse-spark-open" }, "sameCompany": [ "meta-model-api", "muse-code", "muse-glimmer-30b" ], "family": { "lab": "Meta", "family": "Muse", "url": "https://www.anchorterminal.com/open-weights/#meta", "others": [ { "name": "Muse Glimmer 30B", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "params": "30B (dense, multimodal)", "listing": "https://www.anchorterminal.com/tools/muse-glimmer-30b" }, { "name": "Muse Glimmer 30B assistant (DFlash drafter)", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B-assistant", "params": "2.56B (Hub safetensors count; the card states no figure)" } ], "olderCount": 50, "unrestrictedCopies": 1 }, "area": "models", "provenance": { "legalEntity": "Meta Platforms, Inc.", "domain": "meta.ai", "domainRegistered": "2017-12-16", "endpointOnVendorDomain": null, "terms": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B/blob/main/LICENSE.md", "privacy": "", "statusPage": "", "changelog": "", "securityTxt": "none", "checked": "2026-11-04", "notes": [ "The licence names Meta Platforms, Inc. as copyright holder, and defines Meta as Meta Platforms Ireland Limited for licensees in the European region and Meta Platforms, Inc. elsewhere. The model card names the same two as model provider.", "The licence is the document that governs use of the weights, so it is recorded as the terms. LICENSE.md matches the text shown on the gated access form word for word. The Usage Policy is its Exhibit A. The separate USE_POLICY.md that Meta's docs link answered 401 without access and wasn't read.", "No privacy policy governs the model, because the owner runs it and no input reaches Meta. The access form says its details are handled under the Meta Privacy Policy on www.facebook.com, whose robots.txt asks readers like ours not to fetch it, so it wasn't read.", "The domain is meta.ai, where Meta's developer docs (dev.meta.ai) and research post (research.meta.ai) sit, the same as the meta-model-api and muse-code listings. RDAP gives a registration date of 16 December 2017.", "No hosted endpoint from Meta, so no status page. No changelog for the model was found, and the Hugging Face commit history needs access to read.", "dev.meta.ai/.well-known/security.txt and research.meta.ai/.well-known/security.txt returned 404 on 15 October 2026.", "WebFetch read the research post. It refused the Hugging Face model card as outside its allowed set, so the card, LICENSE.md, the repository and collection metadata, the discussions, the evaluation methodology PDF, Meta's docs pages and the providers' model pages were read with curl.", "Re-read on 4 November 2026 from the Hugging Face model pages, which show both Meta repositories gated with manual review and last changed on 15 October 2026. Meta's Get the model page now also lists a signed-in download from dev.meta.ai." ], "score": 55 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/muse-spark-open.json", "live": { "slug": "muse-spark-open", "securityTxt": { "url": "https://meta.ai/.well-known/security.txt", "state": "none", "checkedAt": "2026-11-04T15:52:55.744618885Z" }, "llmsTxt": { "url": "https://dev.meta.ai/llms.txt", "ok": true, "status": 200, "checkedAt": "2026-11-04T15:23:40.027755541Z" }, "domain": { "domain": "meta.ai", "registered": "2017-12-16", "source": "https://rdap.identitydigital.services/rdap/domain/meta.ai", "checkedAt": "2026-11-01T14:15:25.518716318Z" }, "pages": [ { "url": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B/blob/main/LICENSE.md", "kind": "terms", "status": 200, "checkedAt": "2026-11-04T08:26:08.480502101Z", "changedAt": "2026-11-04T08:26:08.480502101Z", "fingerprint": "59003271fce2" } ], "models": { "meta/muse-spark-open": { "openrouterId": "meta/muse-spark-open", "tools": true, "structuredOutputs": true, "jsonMode": true, "reasoning": true, "input": [ "audio", "file", "image", "text", "video" ], "promptCaching": true, "contextTokens": 1048576, "maxOutput": 943718, "checkedAt": "2026-11-04T23:49:21.961528872Z" } }, "updatedAt": "2026-11-04T23:49:21.961528872Z" } }, "answer": "Muse Spark Open scores 53.4 (D) on agent readiness against SWE-2 Flash's 36.4 (F), and leads in 4 of 7 scored categories. SWE-2 Flash leads on reliability, security \u0026 auth and transparency \u0026 trust.", "b": { "slug": "swe-2-flash", "name": "SWE-2 Flash", "vendor": "Cognition AI, Inc.", "vendorUrl": "https://cognition.com/blog/swe-2-flash", "kind": "model", "category": "inference", "summary": "SWE-2 Flash is Cognition's small, fast coding model, mid-trained and reinforcement-trained from Thinking Machines' Inkling-Small (276B total, 12B active parameters). It runs inside Devin Desktop and Devin CLI, with no model API of its own and no public weights.", "url": "https://www.anchorterminal.com/tools/swe-2-flash", "markdownUrl": "https://www.anchorterminal.com/tools/swe-2-flash.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/swe-2-flash.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/swe-2-flash.json", "dns": "swe-2-flash.grade.anchorterminal.com", "license": "Proprietary model, used through Devin Desktop and Devin CLI under Cognition's Platform Terms of Service. The weights are not public; Cognition says it is building a programme to share them with industry partners and researchers", "transports": [], "packages": [], "auth": "mixed", "authNotes": "No model API or key of its own. A person signs in to Devin Desktop, or runs `devin auth login` for Devin CLI, through a browser (enterprise users through their identity provider). The CLI then keeps an API token in credentials.toml that by default is persistent and does not expire, and the docs warn that anyone with the file can authenticate as that user. Enterprise admins choose which models a team may use with a model allowlist.", "pricing": "freemium", "pricingNotes": "Free in Devin Desktop and Devin CLI through 30 November 2026, per the launch post. Afterwards it is billed against the plan's daily and weekly quota, then extra usage at API pricing. The post says one-fifth the price of SWE-2, whose list price is $3.00 input, $0.30 cached input and $15.00 output per million tokens, but no SWE-2 Flash figure is published: the docs' cost table loads in the browser and the models pages don't list it yet (checked 2026-11-04). Plans are Free $0, Pro $20 a month, Max $200 a month and Teams $80 a month plus $40 per full seat. The docs say Free tier users only have SWE-1.6, so whether Free reaches SWE-2 Flash is unconfirmed.", "priceSummary": "Freemium", "where": "local", "x402": { "level": "no", "evidence": "No model endpoint to pay. No x402, MPP or L402 on the Devin pricing page, the Devin Desktop models page or the launch post (checked 2026-11-04).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-11-04" }, "docsUrl": "https://docs.devin.ai/desktop/models", "llmsTxt": "https://docs.devin.ai/llms.txt", "capabilities": [ "inference.llm" ], "tags": [ "official", "hosted", "model", "closed-source", "coding", "moe", "cli", "desktop-app", "llms-txt", "status-page", "soc2", "iso27001", "sla", "freemium", "new" ], "lastRelease": "2026-11-04", "graded": true, "anchor": { "graded": true, "score": 36.4, "grade": "F", "agentReady": false, "rank": 1243, "ranked": true, "rankOf": 1350, "categoryRank": 33, "methodology": "0.7", "run": "2026-10-01", "scores": { "ergonomics": 10, "maintenance": 53, "payments": 20, "reliability": 50, "schema": 29, "security": 53, "transparency": 64 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "low", "date": "2026-11-04" }, "negative": -2, "negativeNotes": [ "The launch post says 'On leading evals, SWE-2 Flash outperforms all American open-weight models'. The post's own table, a first-hand source, gives Meta's Muse Spark Open (American, open weights, released 15 October 2026) 54.0 on SWE-Atlas QA against SWE-2 Flash's 50.0, and Meta's own figure for Muse Spark Open is 58. A 'surpasses' claim a first-hand source shows false is a deduction of 2 (research/new-2026-11-04-ripple-swe2/NEWLIST.md, rule 4). The case against: SWE-2 Flash leads every other row it shares with the American open-weight models in the table, and DeepSWE 1.1 is 60.1 to 60.0, so 'on leading evals' can be read as overall. Proposed for the founder to confirm at landing, as xiaomi-ocr-0's was" ], "verdict": "A new small coding model from Cognition that anyone can try free in Devin Desktop and Devin CLI until 30 November 2026, with strong vendor-reported coding scores. It has no model API, no published per-token price, context length or model card, and no public weights, so an agent can reach it only by driving Devin's own CLI.", "bestFor": "People already using Devin Desktop or Devin CLI who want a cheaper, faster model for routine coding work such as automations, reviews and small changes, while it is free in November 2026.", "strengths": [ "Free in Devin Desktop and Devin CLI through 30 November 2026, and selectable per run with `devin --model` or `/model`", "Cognition reports large gains over its base model, Terminal-Bench 4.0 from 0.9 to 20.0 and FrontierCode 2.0 from 3.2 to 24.0, with the full benchmark table in the post", "Runs inside Devin CLI's sandbox, permissions and hooks, and enterprise admins can allow or block it per team", "Cognition holds SOC 2 Type II and ISO/IEC 27001:2022 and publishes an SLA for contracts that include one", "Status page with a Desktop Agent component, and in the 90 days to 4 November no incident names Desktop or the CLI, only a scheduled extension marketplace maintenance on 7 August" ], "weaknesses": [ "No model API, OpenAPI entry or model card. The Devin API's session agents list SWE-2 but not SWE-2 Flash", "No per-token price, context length, output limit or reasoning settings published for it, and the docs' models pages don't list it yet", "The weights are withheld for now, with a partner and researcher programme promised but not yet described", "Customer data may be used for model training by default on self-serve plans, with the opt-out on paid plans only", "The CLI's login token is persistent and doesn't expire by default, and no rate limit figures or retry guidance are published" ], "agentNotes": [ "There is no endpoint to call. Run Devin CLI single-turn with `devin -p` and pick the model with `--model`; the SWE-2 Flash model ID isn't documented yet, so check `/model` in the CLI first", "Don't rely on the short name `swe`. The docs say it resolves to the latest model in the family, which may be SWE-2 rather than SWE-2 Flash", "Plan for billing to start on 1 December 2026. The free period covers Devin Desktop and CLI only, not Devin cloud sessions", "Treat Cognition's benchmark figures as the vendor's. No independent result for SWE-2 Flash was found", "Keep `credentials.toml` private. The token in it doesn't expire by default and carries the signed-in user's access" ], "metrics": { "kind": "remote", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "low", "grade": "F", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 36.4 } ], "editorialScores": { "ergonomics": 10, "maintenance": 53, "payments": 20, "reliability": 50, "schema": 29, "security": 53, "transparency": 56 }, "provenanceScore": 72, "checked": { "categories": { "ergonomics": "2026-11-04", "maintenance": "2026-11-04", "payments": "2026-11-04", "reliability": "2026-11-04", "schema": "2026-11-04", "security": "2026-11-04", "transparency": "2026-11-04" }, "last": "2026-11-04", "oldest": "2026-11-04", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "Install Devin CLI or Devin Desktop from https://devin.ai/download, sign in with `devin auth login`, then choose SWE-2 Flash with `/model`" }, "letme": { "capability": "https://letme.dev/inference.llm", "tool": "https://letme.dev/swe-2-flash" }, "area": "models", "provenance": { "legalEntity": "Cognition AI, Inc.", "domain": "cognition.com", "domainRegistered": "", "domainNote": "Not looked up: RDAP isn't requested under the 1 November 2026 rules. The product runs on devin.ai, registered 2022-12-06 per the devin listing's RDAP check of 8 October 2026.", "endpointOnVendorDomain": true, "terms": "https://cognition.com/legal/platform-terms-of-service", "privacy": "https://cognition.com/legal/privacy-policy", "statusPage": "https://www.devinstatus.com", "changelog": "https://docs.devin.ai/release-notes/overview", "securityTxt": "none", "checked": "2026-11-04", "notes": [ "The launch post is on cognition.com, signed by the Cognition Team, and the Devin docs, pricing page and status page link the same company. The platform terms (30 June 2026) name Cognition AI, Inc.", "The website terms of use (13 May 2024) and platform terms were read for automated-access limits and have none. cognition.com's robots.txt disallows only /downloads/, devin.ai's only /top-secret and /assets/downloads/, docs.devin.ai's /cdn-cgi/ and /_next/ (with Content-Signal ai-input=yes), and www.devinstatus.com's /api/ and /embed/.", "security.txt is none per the devin listing's 8 October check, not re-read. The docs give a security address for vulnerability reports.", "No SWE-2 Flash entry in the Devin release notes, whose newest is 2 November 2026." ], "score": 72 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/swe-2-flash.json", "live": { "slug": "swe-2-flash", "vendorStatus": { "page": "https://www.devinstatus.com", "indicator": "none", "summary": "All Systems Operational", "checkedAt": "2026-11-05T01:27:53.220664573Z" }, "updatedAt": "2026-11-05T01:27:53.220664573Z" } }, "edges": [ { "by": 59, "category": "ergonomics", "leads": "muse-spark-open", "name": "Agent ergonomics", "reasons": { "a": "Model reading of the checklist (tool use, structured output, caching, context, throughput, defaults and SDKs, errors). OpenAI-format tools and parallel calls come back as `tool_calls`, and the template supports namespaced tools and deferred loading through a `tool_search` tool. A forced `tool_choice` is not documented (17 of 20). JSON mode and JSON Schema come from each runtime's guided decoding, which Meta's docs show for TensorRT-LLM only (8 of 15). Prefix caching is the runtime's. Meta's SGLang command turns on hierarchical caching, and Dynamo routes each request to the worker holding its cached context (8 of 10). 1,048,576-token context, with sliding-window attention on 46 of 62 layers holding KV memory down (15 of 15). Draft models for speculative decoding in SGLang and vLLM, and no batch mode of its own (7 of 10). Any OpenAI SDK works against a local server, but reasoning can't be turned off, defaults to the most expensive level, and needs streaming, a `max_tokens` budget for reasoning and the right stop tokens (6 of 15). The template raises named errors for an unsupported reasoning effort, tool arguments passed as a string and a tool message with no name. Server errors are the runtime's (8 of 15). 17 + 8 + 8 + 15 + 7 + 6 + 8 = 69.", "b": "Model reading (tool use, structured output, caching, context, batch, SDKs). Tool use is Devin's harness, not something a caller configures; no context length, output limit or cache price is published for SWE-2 Flash (5 of 25). No documented output-size or reasoning controls for this model; SWE-2's effort levels aren't stated for Flash (0 of 20). No model-level error documentation (0 of 20). No idempotency or retry guidance (0 of 20). One flag or picker selects it and `devin -p` runs a single turn and prints to stdout, but there is no SDK or direct endpoint (5 of 15). 5 + 0 + 0 + 0 + 5 = 10." }, "scores": { "a": 69, "b": 10 } }, { "by": 32, "category": "schema", "leads": "muse-spark-open", "name": "Schema \u0026 documentation", "reasons": { "a": "Read for a model the owner serves. No specification of its own. The contract is the Jinja chat template, which renders OpenAI-format tools with `additionalProperties: false` added and rejects bad values, plus the OpenAI-compatible servers in vLLM and SGLang with the `muse` parsers (10 of 25). dev.meta.ai/llms.txt lists the Muse Spark Open overview and seven pages under it, plus three cookbook pages, each with a Markdown copy (10). The card lists seven intended uses and six limitations, the docs say when to pick BF16 or NVFP4 and which reasoning strength suits which task, and the Usage Policy states what is out of scope (17 of 20). `reasoning_effort` is an enum of four values and anything else returns an error, and tools take JSON Schema. Structured output is documented only for TensorRT-LLM's `response_format`, not in the prompting guide or for vLLM and SGLang (11 of 15). curl and Python examples for every runtime and hosted provider, the rendered prompt for each case, and the template's own error messages, but no error reference for a served model (11 of 15). One release, with no changelog or version scheme beyond the licence's version line, Muse Spark Open (2 of 15). 10 + 10 + 17 + 11 + 11 + 2 = 61.", "b": "Model reading. No model API, so no OpenAPI entry; the Devin v3 OpenAPI file lists SWE-2 agents only, and no model card was found (0). docs.devin.ai serves llms.txt and Markdown copies of every page (10). The launch post says what it is for (automations, code review, feature development, documentation) and the CLI docs say when to pick a SWE model, but the models pages don't list SWE-2 Flash yet (6 of 20). No documented model ID, reasoning settings, context or output limits to type (0 of 15). The CLI pages show `--model` and `/model` examples, with no model-specific errors (3 of 15). Named, dated model versions on Cognition's blog and public release notes, but no release note for SWE-2 Flash yet; the newest is 2 November (10 of 15). 0 + 10 + 6 + 0 + 3 + 10 = 29." }, "scores": { "a": 61, "b": 29 } }, { "by": 30, "category": "payments", "leads": "muse-spark-open", "name": "Payments \u0026 pricing", "reasons": { "a": "Self-hosted rule. Nothing to buy from Meta, and the licence grants use free of charge. The Model as a Service clause asks for a separate agreement only from licensee groups above $20M gross revenue in 12 months (20). Free to run with no card (20). Meta's own weights need a manually reviewed Hugging Face form with legal name and date of birth, or, since the 4 November re-read of Get the model, a signed-in download from Meta's own page, so an agent can't get them alone. The ungated GGUF quantisations from ggml-org and Unsloth that Meta's docs point to download without approval (the Unsloth repository is ungated on 4 November), so part of the line (10 of 20). No payment protocol (0). Hosted copies at $1.25 input and $4.25 output per million tokens need each provider's own account and are not graded. 20 + 20 + 10 + 0 = 50.", "b": "Published rubric. No payment protocol (0). Plan prices are public, but no per-token price for SWE-2 Flash is on a readable page; the docs' cost table loads in the browser and the post gives only 'one-fifth the price' of SWE-2 (10). Free through 30 November 2026 on Devin Desktop and CLI, but the docs say Free tier users only have SWE-1.6, so whether the Free plan reaches it, and without a card, isn't confirmed; the in-between value used on the devin listing (10). A person signs in through a browser (0). 0 + 10 + 10 + 0 = 20." }, "scores": { "a": 50, "b": 20 } }, { "by": 11, "category": "transparency", "leads": "swe-2-flash", "name": "Transparency \u0026 trust", "reasons": { "a": "Weights are downloadable under the Muse Model Licence, which is clear and broad in its grant but not an OSI licence. It adds a display requirement, a separate agreement before Model as a Service by a licensee group above $20M gross revenue, and a Usage Policy with field-of-use limits (military, nuclear, espionage, critical infrastructure, heavy machinery) and an age limit of 18. It has no regional exclusion and no governing-law clause. Training code and data aren't released (18 of 30). Inputs stay on the owner's hardware, and Meta's docs say self-hosting keeps 'every prompt and response' inside the owner's infrastructure. The access form's details, and a download from Meta's own page after signing in, go to Meta under its general privacy policy, which we didn't read, and the training data is described in outline only (20 of 30). No deprecation policy, dated notices or versioning found (0 of 20). The Get the model page says everything runs offline once the weights land, and safetensors weights carry no code. No other telemetry statement, and Meta's download form records name, date of birth, country and IP location (12 of 20). 18 + 20 + 0 + 12 = 50.", "b": "Editorial half. Closed model under Cognition's platform terms; the weights aren't public and the sharing programme isn't described yet (15). Terms, privacy policy, security page and DPA agree, with training on self-serve data by default and no retention periods, as read for the devin listing (18). The API versioning policy covers API endpoints, not models, and no model deprecation notice was found (5 of 20). Sub-processors listed with purpose and country, per the devin listing's 8 October check (18). 15 + 18 + 5 + 18 = 56." }, "scores": { "a": 53, "b": 64 } }, { "by": 9, "category": "security", "leads": "swe-2-flash", "name": "Security \u0026 auth", "reasons": { "a": "Read as software the owner runs. No account or key to run it. Meta's weights need manual approval on Hugging Face and an access token, and server authentication is the owner's to add (12 of 30). An agent model that calls whatever tools the operator gives it. The card recommends human-in-the-loop confirmation for irreversible actions and Meta reports agentic-risk evaluations, but nothing in the model or template enforces an approval (8 of 20). The chat template neutralises special-token sequences and media placeholders in caller text and wraps tool results in `\u003ctool_output\u003e`, and the card advises protecting internal assistant messages and layering defences against adversarial input. No indirect prompt-injection results are published (10 of 15). No call log of its own. Visibility depends on the serving stack (5 of 15). Weights are safetensors that load through Transformers' own classes from 5.19, and Meta published cyber, chemical, biological and misalignment evaluations. dev.meta.ai/.well-known/security.txt answers 404 (re-read on 4 November), research.meta.ai per the 15 October check, and no security route is on the card or in the docs, only museoss@meta.com and the Hugging Face discussions (9 of 20). 12 + 8 + 10 + 5 + 9 = 44.", "b": "Read as a hosted model reached through Devin's local clients. The CLI signs in through a browser, or the enterprise identity provider, and stores an API token in credentials.toml that by default is persistent and does not expire and carries the user's access, with no scopes found for it (10 of 30). Devin CLI documents an OS-level sandbox with network filtering, hooks that can block tool calls, and enterprise model allowlists and team settings (14 of 20). Guardrails against prompt injection are Enterprise only and cover user messages, as on the devin listing (8 of 15). Enterprise analytics count tool calls and consumption by model and user; no per-user audit log for the CLI was found (8 of 15). SOC 2 Type II and ISO/IEC 27001:2022, a disclosure address, no bug bounty and no security.txt, per the devin listing (13 of 20). Training: self-serve customer data may train Cognition's models by default, with an opt-out on paid plans that also turns on zero data retention at model providers. Cognition withheld the weights over cyber misuse, which is recorded as a fact, not scored. 10 + 14 + 8 + 8 + 13 = 53." }, "scores": { "a": 44, "b": 53 } }, { "by": 8, "category": "reliability", "leads": "swe-2-flash", "name": "Reliability", "reasons": { "a": "Read with the local-software lines, since the owner serves the weights. Meta's weights come from two Hugging Face repositories gated with manual review. The docs state Python 3.10 or later, Transformers 5.19 or later, Docker images for SGLang and vLLM and the GPUs for each checkpoint (4 × B300 for NVFP4, 8 × B300 for BF16), but SGLang's image is a dev tag and TensorRT-LLM's support sits on a feature branch (13 of 20). No CI or test suite of Meta's for the model was found. github.com/meta-models answered 403 to our fetcher, so its repositories are unchecked (0 of 25). On 4 November the BF16 repository's discussions page lists 13 open threads from 15 October to 1 November, 10 discussions and 3 pull requests (one a draft). A script counted 29 replies across the seven threads with replies, six discussions and the draft pull request, none from the repository owner or a Meta member, and their text wasn't read, so crash reports weren't separated from other posts (15 of 25, held from 15 October and flagged). No changelog, version tags or revision notes, and the commit history needs gated access (2 of 15). Released as a final checkpoint with no preview label from Meta, and both repositories unchanged since 15 October, though runtime support ships as launch-specific images and a branch, and Groq lists it as a preview (12 of 15). 13 + 0 + 15 + 2 + 12 = 42.", "b": "Model reading, for the surface it runs on, Devin Desktop and Devin CLI. Status page at www.devinstatus.com (Atlassian Statuspage) with Desktop Agent and Desktop Tab components besides the cloud ones, read on 4 November 2026 (20). The history feed for the 90 days from 6 August lists 'Devin unavailable' for 20 minutes on 13 August with no component named, read as across the board and a major as on the devin listing; Devin Cloud unavailable on 13 August and the web app down on 24 September name cloud surfaces this model doesn't use; the rest are degraded-performance incidents (new sessions on 3 September, 'multiple services' on 17 September, Ask Devin on 26 October, automation triggers on 4 November) and session queuing on 10 and 11 August, read as minor and a scheduled Devin Desktop extension marketplace upgrade on 7 August, read as maintenance. One major (10). No rate limit figures for the model or the CLI; usage is a daily and weekly quota with no numbers (0). No 429, backoff or retry guidance for this surface (0). Cognition publishes an SLA (25 February 2026) with uptime and response times that applies when included in an order form, read as a contractual SLA on Enterprise, as on other listings (10). The launch post makes it available today with no preview or beta label (10). 20 + 10 + 0 + 0 + 10 + 10 = 50." }, "scores": { "a": 42, "b": 50 } } ], "facts": [ { "a": "Model API", "b": "Model API", "name": "Kind" }, { "a": "Meta", "b": "Cognition AI, Inc.", "name": "Vendor" }, { "a": "no (local only)", "b": "no (local only)", "name": "Hosted endpoint" }, { "a": "HTTP", "b": "", "name": "Transports" }, { "a": "None", "b": "OAuth or key", "name": "Auth" }, { "a": "Free", "b": "Freemium", "name": "Pricing" }, { "a": "no", "b": "no", "name": "x402" }, { "a": "Muse Model Licence (source-available weights, not an OSI licence), free of charge with an attribution notice, a separate agreement with Meta before Model as a Service by a licensee group above $20M gross revenue, and the Muse Spark Open Usage Policy", "b": "Proprietary model, used through Devin Desktop and Devin CLI under Cognition's Platform Terms of Service. The weights are not public; Cognition says it is building a programme to share them with industry partners and researchers", "name": "Licence" }, { "a": "no", "b": "no", "name": "Read-only variant documented" }, { "a": "yes", "b": "yes", "name": "llms.txt" }, { "a": "2026-10-15", "b": "2026-11-04", "name": "Last release" }, { "a": "2026-10-15", "b": "2026-06-30", "name": "Terms last updated" }, { "a": "no document linked", "b": "2026-03-09", "name": "Privacy policy last updated" }, { "a": "not found in the text", "b": "yes, with an opt-out", "name": "Customer content may train models" }, { "a": "not found in the text", "b": "not found in the text", "name": "Terms restrict automated access" }, { "a": "not found in the text", "b": "yes", "name": "Terms restrict benchmarking" }, { "a": "not found in the text", "b": "not found in the text", "name": "Terms or service can change without notice" }, { "a": "not found in the text", "b": "yes", "name": "Arbitration or class-action waiver" } ], "faq": [ { "answer": "Muse Spark Open scores 53.4 (D) on agent readiness against SWE-2 Flash's 36.4 (F), and leads in 4 of 7 scored categories. SWE-2 Flash leads on reliability, security \u0026 auth and transparency \u0026 trust.", "question": "Which is better for AI agents, Muse Spark Open or SWE-2 Flash?" }, { "answer": "Muse Spark Open needs no key. SWE-2 Flash takes an API key or an OAuth sign-in.", "question": "Do Muse Spark Open and SWE-2 Flash need an API key?" }, { "answer": "No hosted endpoint is listed for Muse Spark Open. No hosted endpoint is listed for SWE-2 Flash.", "question": "Can an agent call Muse Spark Open and SWE-2 Flash without installing anything?" } ], "goodFor": [ { "aheadOn": [ "Schema \u0026 documentation, 61 against 29", "Agent ergonomics, 69 against 10", "Payments \u0026 pricing, 50 against 20", "Maintenance \u0026 community, 60 against 53" ], "also": [ "No key needed to call it" ], "goodFor": "A team with a four-GPU Blackwell node, or two to four Mac Studios, that wants a 1M-token multimodal coding and agent model on its own hardware with no prompts leaving it.", "slug": "muse-spark-open", "watchFor": "The NVFP4 checkpoint (about 537 GB) serves on four B300 GPUs and the BF16 one (about 1.65 TB) on eight" }, { "aheadOn": [ "Reliability, 50 against 42", "Security \u0026 auth, 53 against 44", "Transparency \u0026 trust, 64 against 53" ], "also": null, "goodFor": "People already using Devin Desktop or Devin CLI who want a cheaper, faster model for routine coding work such as automations, reviews and small changes, while it is free in November 2026.", "slug": "swe-2-flash", "watchFor": "No model API, OpenAPI entry or model card. The Devin API's session agents list SWE-2 but not SWE-2 Flash" } ], "history": { "muse-spark-open": [ { "causes": [ "method" ], "date": "2026-10-30", "delta": -0.1, "from": 53.5, "fromGrade": "D", "to": 53.4, "toGrade": "D" }, { "causes": [ "evidence" ], "date": "2026-10-28", "delta": 0, "from": 53.5, "fromGrade": "D", "to": 53.5, "toGrade": "D" }, { "causes": [ "method" ], "date": "2026-10-27", "delta": 0.1, "from": 53.4, "fromGrade": "D", "to": 53.5, "toGrade": "D" } ], "swe-2-flash": null }, "job": { "capability": "inference.llm", "name": "LLM inference" }, "others": [ { "json": "https://www.anchorterminal.com/compare/anthropic-api-vs-muse-spark-open.json", "title": "Claude API vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/anthropic-api-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/anthropic-api-vs-swe-2-flash.json", "title": "Claude API vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/anthropic-api-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/antseed-vs-muse-spark-open.json", "title": "Antseed vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/antseed-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/antseed-vs-swe-2-flash.json", "title": "Antseed vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/antseed-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/blockrun-ai-vs-muse-spark-open.json", "title": "BlockRun.AI vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/blockrun-ai-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/blockrun-ai-vs-swe-2-flash.json", "title": "BlockRun.AI vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/blockrun-ai-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/cerebras-inference-vs-muse-spark-open.json", "title": "Cerebras Inference vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/cerebras-inference-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/cerebras-inference-vs-swe-2-flash.json", "title": "Cerebras Inference vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/cerebras-inference-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/cloudflare-ai-gateway-vs-muse-spark-open.json", "title": "Cloudflare AI Gateway vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/cloudflare-ai-gateway-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/cloudflare-ai-gateway-vs-swe-2-flash.json", "title": "Cloudflare AI Gateway vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/cloudflare-ai-gateway-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/cloudflare-workers-ai-vs-muse-spark-open.json", "title": "Cloudflare Workers AI vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/cloudflare-workers-ai-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/cloudflare-workers-ai-vs-swe-2-flash.json", "title": "Cloudflare Workers AI vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/cloudflare-workers-ai-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/cohere-chat-vs-muse-spark-open.json", "title": "Cohere Chat API (Command models) vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/cohere-chat-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/cohere-chat-vs-swe-2-flash.json", "title": "Cohere Chat API (Command models) vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/cohere-chat-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/deepinfra-vs-muse-spark-open.json", "title": "DeepInfra vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/deepinfra-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/deepinfra-vs-swe-2-flash.json", "title": "DeepInfra vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/deepinfra-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/deepseek-api-vs-muse-spark-open.json", "title": "DeepSeek API vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/deepseek-api-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/deepseek-api-vs-swe-2-flash.json", "title": "DeepSeek API vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/deepseek-api-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/elorian-glimpse-vs-muse-spark-open.json", "title": "Glimpse vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/elorian-glimpse-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/elorian-glimpse-vs-swe-2-flash.json", "title": "Glimpse vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/elorian-glimpse-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/fireworks-inference-vs-muse-spark-open.json", "title": "Fireworks AI Inference vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/fireworks-inference-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/fireworks-inference-vs-swe-2-flash.json", "title": "Fireworks AI Inference vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/fireworks-inference-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/gemini-api-vs-muse-spark-open.json", "title": "Gemini Developer API vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/gemini-api-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/gemini-api-vs-swe-2-flash.json", "title": "Gemini Developer API vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/gemini-api-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/groq-vs-muse-spark-open.json", "title": "GroqCloud vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/groq-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/groq-vs-swe-2-flash.json", "title": "GroqCloud vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/groq-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/jais-3-vs-muse-spark-open.json", "title": "Jais 3 vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/jais-3-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/jais-3-vs-swe-2-flash.json", "title": "Jais 3 vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/jais-3-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/meta-model-api-vs-muse-spark-open.json", "title": "Meta Model API vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/meta-model-api-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/meta-model-api-vs-swe-2-flash.json", "title": "Meta Model API vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/meta-model-api-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/minimax-api-vs-muse-spark-open.json", "title": "MiniMax API vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/minimax-api-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/minimax-api-vs-swe-2-flash.json", "title": "MiniMax API vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/minimax-api-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/mistral-api-vs-muse-spark-open.json", "title": "Mistral AI API vs Muse Spark Open", "url": "https://www.anchorterminal.com/compare/mistral-api-vs-muse-spark-open" }, { "json": "https://www.anchorterminal.com/compare/mistral-api-vs-swe-2-flash.json", "title": "Mistral AI API vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/mistral-api-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-nemotron-3-5-super-vl.json", "title": "Muse Spark Open vs Nemotron 3.5 Super VL", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-nemotron-3-5-super-vl" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-novita-ai.json", "title": "Muse Spark Open vs Novita AI", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-novita-ai" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-openai-api.json", "title": "Muse Spark Open vs OpenAI API", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-openai-api" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-openrouter.json", "title": "Muse Spark Open vs OpenRouter", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-openrouter" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-prism-inference.json", "title": "Muse Spark Open vs Prism Inference", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-prism-inference" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-ramp-router.json", "title": "Muse Spark Open vs Router.com (Ramp Router)", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-ramp-router" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-reflection-beam.json", "title": "Muse Spark Open vs Beam 1.0", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-reflection-beam" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-routerplus.json", "title": "Muse Spark Open vs RouterPlus", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-routerplus" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-sambanova.json", "title": "Muse Spark Open vs SambaCloud", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-sambanova" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-siliconflow.json", "title": "Muse Spark Open vs SiliconFlow", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-siliconflow" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-simplismart.json", "title": "Muse Spark Open vs Simplismart", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-simplismart" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-stepfun-api.json", "title": "Muse Spark Open vs StepFun API", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-stepfun-api" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-tencent-tokenhub.json", "title": "Muse Spark Open vs Tencent Cloud TokenHub", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-tencent-tokenhub" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-together-inference.json", "title": "Muse Spark Open vs Together AI Inference", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-together-inference" }, { "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-vercel-ai-gateway.json", "title": "Muse Spark Open vs Vercel AI Gateway", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-vercel-ai-gateway" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-super-vl-vs-swe-2-flash.json", "title": "Nemotron 3.5 Super VL vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-super-vl-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/novita-ai-vs-swe-2-flash.json", "title": "Novita AI vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/novita-ai-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/openai-api-vs-swe-2-flash.json", "title": "OpenAI API vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/openai-api-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/openrouter-vs-swe-2-flash.json", "title": "OpenRouter vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/openrouter-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/prism-inference-vs-swe-2-flash.json", "title": "Prism Inference vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/prism-inference-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/ramp-router-vs-swe-2-flash.json", "title": "Router.com (Ramp Router) vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/ramp-router-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/reflection-beam-vs-swe-2-flash.json", "title": "Beam 1.0 vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/reflection-beam-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/routerplus-vs-swe-2-flash.json", "title": "RouterPlus vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/routerplus-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/sambanova-vs-swe-2-flash.json", "title": "SambaCloud vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/sambanova-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/siliconflow-vs-swe-2-flash.json", "title": "SiliconFlow vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/siliconflow-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/simplismart-vs-swe-2-flash.json", "title": "Simplismart vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/simplismart-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/stepfun-api-vs-swe-2-flash.json", "title": "StepFun API vs SWE-2 Flash", "url": "https://www.anchorterminal.com/compare/stepfun-api-vs-swe-2-flash" }, { "json": "https://www.anchorterminal.com/compare/swe-2-flash-vs-tencent-tokenhub.json", "title": "SWE-2 Flash vs Tencent Cloud TokenHub", "url": "https://www.anchorterminal.com/compare/swe-2-flash-vs-tencent-tokenhub" }, { "json": "https://www.anchorterminal.com/compare/swe-2-flash-vs-together-inference.json", "title": "SWE-2 Flash vs Together AI Inference", "url": "https://www.anchorterminal.com/compare/swe-2-flash-vs-together-inference" }, { "json": "https://www.anchorterminal.com/compare/swe-2-flash-vs-vercel-ai-gateway.json", "title": "SWE-2 Flash vs Vercel AI Gateway", "url": "https://www.anchorterminal.com/compare/swe-2-flash-vs-vercel-ai-gateway" } ], "panel": null, "scores": [ { "by": 8, "edge": "swe-2-flash", "key": "reliability", "muse-spark-open": 42, "name": "Reliability", "swe-2-flash": 50, "weight": 16 }, { "key": "performance", "name": "Performance", "pending": true, "weight": 10 }, { "by": 32, "edge": "muse-spark-open", "key": "schema", "muse-spark-open": 61, "name": "Schema \u0026 documentation", "swe-2-flash": 29, "weight": 13 }, { "by": 59, "edge": "muse-spark-open", "key": "ergonomics", "muse-spark-open": 69, "name": "Agent ergonomics", "swe-2-flash": 10, "weight": 13 }, { "by": 9, "edge": "swe-2-flash", "key": "security", "muse-spark-open": 44, "name": "Security \u0026 auth", "swe-2-flash": 53, "weight": 14 }, { "by": 30, "edge": "muse-spark-open", "key": "payments", "muse-spark-open": 50, "name": "Payments \u0026 pricing", "swe-2-flash": 20, "weight": 10 }, { "key": "tasks", "name": "Task success", "pending": true, "weight": 10 }, { "by": 7, "edge": "muse-spark-open", "key": "maintenance", "muse-spark-open": 60, "name": "Maintenance \u0026 community", "swe-2-flash": 53, "weight": 7 }, { "by": 11, "edge": "swe-2-flash", "key": "transparency", "muse-spark-open": 53, "name": "Transparency \u0026 trust", "swe-2-flash": 64, "weight": 7 } ], "summary": "Muse Spark Open scores 53.4 (D) on agent readiness against SWE-2 Flash's 36.4 (F), and leads in 4 of 7 scored categories. SWE-2 Flash leads on reliability, security \u0026 auth and transparency \u0026 trust. Both do llm inference.", "tasks": null, "tutorials": [], "verdicts": { "muse-spark-open": "Muse Spark Open runs on the owner's hardware in seven runtimes, with OpenAI-compatible servers in vLLM and SGLang, standard tool calls, a 1M-token context and Meta's own serving docs. Serving the 537 GB NVFP4 checkpoint takes four B300 GPUs, Meta's download is gated by manual review, and the licence adds attribution and a revenue threshold for model resellers.", "swe-2-flash": "A new small coding model from Cognition that anyone can try free in Devin Desktop and Devin CLI until 30 November 2026, with strong vendor-reported coding scores. It has no model API, no published per-token price, context length or model card, and no public weights, so an agent can reach it only by driving Devin's own CLI." } }, "kind": "anchor.page", "links": { "api": "https://www.anchorterminal.com/api/v1/index.json", "html": "https://www.anchorterminal.com/compare/muse-spark-open-vs-swe-2-flash", "json": "https://www.anchorterminal.com/compare/muse-spark-open-vs-swe-2-flash.json", "llms": "https://www.anchorterminal.com/llms.txt", "markdown": "https://www.anchorterminal.com/compare/muse-spark-open-vs-swe-2-flash.md", "mcp": "https://www.anchorterminal.com/mcp", "slim": "https://www.anchorterminal.com/compare/muse-spark-open-vs-swe-2-flash.min.md" }, "markdown": "Muse Spark Open scores 53.4 (D) on agent readiness against SWE-2 Flash's 36.4 (F), and leads in 4 of 7 scored categories. SWE-2 Flash leads on reliability, security \u0026 auth and transparency \u0026 trust. Both do llm inference.\n\n- Muse Spark Open: grade D, 53.4/100, rank #845 of 1350. Markdown https://www.anchorterminal.com/tools/muse-spark-open.md · JSON https://www.anchorterminal.com/api/v1/tools/muse-spark-open.json\n- SWE-2 Flash: grade F, 36.4/100, rank #1243 of 1350. Markdown https://www.anchorterminal.com/tools/swe-2-flash.md · JSON https://www.anchorterminal.com/api/v1/tools/swe-2-flash.json\n- Best model APIs and inference for AI agents: https://www.anchorterminal.com/best/inference/index.md\n- All 562 models comparisons: https://www.anchorterminal.com/compare/inference/index.md\n\n## Which one, for what\n\n### Muse Spark Open (D)\n\nGood for: A team with a four-GPU Blackwell node, or two to four Mac Studios, that wants a 1M-token multimodal coding and agent model on its own hardware with no prompts leaving it.\n\nAhead on:\n- Schema \u0026 documentation, 61 against 29\n- Agent ergonomics, 69 against 10\n- Payments \u0026 pricing, 50 against 20\n- Maintenance \u0026 community, 60 against 53\n\nAlso in its favour:\n- No key needed to call it\n\nWatch for: The NVFP4 checkpoint (about 537 GB) serves on four B300 GPUs and the BF16 one (about 1.65 TB) on eight\n\n### SWE-2 Flash (F)\n\nGood for: People already using Devin Desktop or Devin CLI who want a cheaper, faster model for routine coding work such as automations, reviews and small changes, while it is free in November 2026.\n\nAhead on:\n- Reliability, 50 against 42\n- Security \u0026 auth, 53 against 44\n- Transparency \u0026 trust, 64 against 53\n\nWatch for: No model API, OpenAPI entry or model card. The Devin API's session agents list SWE-2 but not SWE-2 Flash\n\n\n## Why the edge\n\n### Agent ergonomics: Muse Spark Open +59\n\n- Muse Spark Open (69): Model reading of the checklist (tool use, structured output, caching, context, throughput, defaults and SDKs, errors). OpenAI-format tools and parallel calls come back as `tool_calls`, and the template supports namespaced tools and deferred loading through a `tool_search` tool. A forced `tool_choice` is not documented (17 of 20). JSON mode and JSON Schema come from each runtime's guided decoding, which Meta's docs show for TensorRT-LLM only (8 of 15). Prefix caching is the runtime's. Meta's SGLang command turns on hierarchical caching, and Dynamo routes each request to the worker holding its cached context (8 of 10). 1,048,576-token context, with sliding-window attention on 46 of 62 layers holding KV memory down (15 of 15). Draft models for speculative decoding in SGLang and vLLM, and no batch mode of its own (7 of 10). Any OpenAI SDK works against a local server, but reasoning can't be turned off, defaults to the most expensive level, and needs streaming, a `max_tokens` budget for reasoning and the right stop tokens (6 of 15). The template raises named errors for an unsupported reasoning effort, tool arguments passed as a string and a tool message with no name. Server errors are the runtime's (8 of 15). 17 + 8 + 8 + 15 + 7 + 6 + 8 = 69.\n- SWE-2 Flash (10): Model reading (tool use, structured output, caching, context, batch, SDKs). Tool use is Devin's harness, not something a caller configures; no context length, output limit or cache price is published for SWE-2 Flash (5 of 25). No documented output-size or reasoning controls for this model; SWE-2's effort levels aren't stated for Flash (0 of 20). No model-level error documentation (0 of 20). No idempotency or retry guidance (0 of 20). One flag or picker selects it and `devin -p` runs a single turn and prints to stdout, but there is no SDK or direct endpoint (5 of 15). 5 + 0 + 0 + 0 + 5 = 10.\n\n### Schema \u0026 documentation: Muse Spark Open +32\n\n- Muse Spark Open (61): Read for a model the owner serves. No specification of its own. The contract is the Jinja chat template, which renders OpenAI-format tools with `additionalProperties: false` added and rejects bad values, plus the OpenAI-compatible servers in vLLM and SGLang with the `muse` parsers (10 of 25). dev.meta.ai/llms.txt lists the Muse Spark Open overview and seven pages under it, plus three cookbook pages, each with a Markdown copy (10). The card lists seven intended uses and six limitations, the docs say when to pick BF16 or NVFP4 and which reasoning strength suits which task, and the Usage Policy states what is out of scope (17 of 20). `reasoning_effort` is an enum of four values and anything else returns an error, and tools take JSON Schema. Structured output is documented only for TensorRT-LLM's `response_format`, not in the prompting guide or for vLLM and SGLang (11 of 15). curl and Python examples for every runtime and hosted provider, the rendered prompt for each case, and the template's own error messages, but no error reference for a served model (11 of 15). One release, with no changelog or version scheme beyond the licence's version line, Muse Spark Open (2 of 15). 10 + 10 + 17 + 11 + 11 + 2 = 61.\n- SWE-2 Flash (29): Model reading. No model API, so no OpenAPI entry; the Devin v3 OpenAPI file lists SWE-2 agents only, and no model card was found (0). docs.devin.ai serves llms.txt and Markdown copies of every page (10). The launch post says what it is for (automations, code review, feature development, documentation) and the CLI docs say when to pick a SWE model, but the models pages don't list SWE-2 Flash yet (6 of 20). No documented model ID, reasoning settings, context or output limits to type (0 of 15). The CLI pages show `--model` and `/model` examples, with no model-specific errors (3 of 15). Named, dated model versions on Cognition's blog and public release notes, but no release note for SWE-2 Flash yet; the newest is 2 November (10 of 15). 0 + 10 + 6 + 0 + 3 + 10 = 29.\n\n### Payments \u0026 pricing: Muse Spark Open +30\n\n- Muse Spark Open (50): Self-hosted rule. Nothing to buy from Meta, and the licence grants use free of charge. The Model as a Service clause asks for a separate agreement only from licensee groups above $20M gross revenue in 12 months (20). Free to run with no card (20). Meta's own weights need a manually reviewed Hugging Face form with legal name and date of birth, or, since the 4 November re-read of Get the model, a signed-in download from Meta's own page, so an agent can't get them alone. The ungated GGUF quantisations from ggml-org and Unsloth that Meta's docs point to download without approval (the Unsloth repository is ungated on 4 November), so part of the line (10 of 20). No payment protocol (0). Hosted copies at $1.25 input and $4.25 output per million tokens need each provider's own account and are not graded. 20 + 20 + 10 + 0 = 50.\n- SWE-2 Flash (20): Published rubric. No payment protocol (0). Plan prices are public, but no per-token price for SWE-2 Flash is on a readable page; the docs' cost table loads in the browser and the post gives only 'one-fifth the price' of SWE-2 (10). Free through 30 November 2026 on Devin Desktop and CLI, but the docs say Free tier users only have SWE-1.6, so whether the Free plan reaches it, and without a card, isn't confirmed; the in-between value used on the devin listing (10). A person signs in through a browser (0). 0 + 10 + 10 + 0 = 20.\n\n### Transparency \u0026 trust: SWE-2 Flash +11\n\n- Muse Spark Open (53): Weights are downloadable under the Muse Model Licence, which is clear and broad in its grant but not an OSI licence. It adds a display requirement, a separate agreement before Model as a Service by a licensee group above $20M gross revenue, and a Usage Policy with field-of-use limits (military, nuclear, espionage, critical infrastructure, heavy machinery) and an age limit of 18. It has no regional exclusion and no governing-law clause. Training code and data aren't released (18 of 30). Inputs stay on the owner's hardware, and Meta's docs say self-hosting keeps 'every prompt and response' inside the owner's infrastructure. The access form's details, and a download from Meta's own page after signing in, go to Meta under its general privacy policy, which we didn't read, and the training data is described in outline only (20 of 30). No deprecation policy, dated notices or versioning found (0 of 20). The Get the model page says everything runs offline once the weights land, and safetensors weights carry no code. No other telemetry statement, and Meta's download form records name, date of birth, country and IP location (12 of 20). 18 + 20 + 0 + 12 = 50.\n- SWE-2 Flash (64): Editorial half. Closed model under Cognition's platform terms; the weights aren't public and the sharing programme isn't described yet (15). Terms, privacy policy, security page and DPA agree, with training on self-serve data by default and no retention periods, as read for the devin listing (18). The API versioning policy covers API endpoints, not models, and no model deprecation notice was found (5 of 20). Sub-processors listed with purpose and country, per the devin listing's 8 October check (18). 15 + 18 + 5 + 18 = 56.\n\n### Security \u0026 auth: SWE-2 Flash +9\n\n- Muse Spark Open (44): Read as software the owner runs. No account or key to run it. Meta's weights need manual approval on Hugging Face and an access token, and server authentication is the owner's to add (12 of 30). An agent model that calls whatever tools the operator gives it. The card recommends human-in-the-loop confirmation for irreversible actions and Meta reports agentic-risk evaluations, but nothing in the model or template enforces an approval (8 of 20). The chat template neutralises special-token sequences and media placeholders in caller text and wraps tool results in `\u003ctool_output\u003e`, and the card advises protecting internal assistant messages and layering defences against adversarial input. No indirect prompt-injection results are published (10 of 15). No call log of its own. Visibility depends on the serving stack (5 of 15). Weights are safetensors that load through Transformers' own classes from 5.19, and Meta published cyber, chemical, biological and misalignment evaluations. dev.meta.ai/.well-known/security.txt answers 404 (re-read on 4 November), research.meta.ai per the 15 October check, and no security route is on the card or in the docs, only museoss@meta.com and the Hugging Face discussions (9 of 20). 12 + 8 + 10 + 5 + 9 = 44.\n- SWE-2 Flash (53): Read as a hosted model reached through Devin's local clients. The CLI signs in through a browser, or the enterprise identity provider, and stores an API token in credentials.toml that by default is persistent and does not expire and carries the user's access, with no scopes found for it (10 of 30). Devin CLI documents an OS-level sandbox with network filtering, hooks that can block tool calls, and enterprise model allowlists and team settings (14 of 20). Guardrails against prompt injection are Enterprise only and cover user messages, as on the devin listing (8 of 15). Enterprise analytics count tool calls and consumption by model and user; no per-user audit log for the CLI was found (8 of 15). SOC 2 Type II and ISO/IEC 27001:2022, a disclosure address, no bug bounty and no security.txt, per the devin listing (13 of 20). Training: self-serve customer data may train Cognition's models by default, with an opt-out on paid plans that also turns on zero data retention at model providers. Cognition withheld the weights over cyber misuse, which is recorded as a fact, not scored. 10 + 14 + 8 + 8 + 13 = 53.\n\n### Reliability: SWE-2 Flash +8\n\n- Muse Spark Open (42): Read with the local-software lines, since the owner serves the weights. Meta's weights come from two Hugging Face repositories gated with manual review. The docs state Python 3.10 or later, Transformers 5.19 or later, Docker images for SGLang and vLLM and the GPUs for each checkpoint (4 × B300 for NVFP4, 8 × B300 for BF16), but SGLang's image is a dev tag and TensorRT-LLM's support sits on a feature branch (13 of 20). No CI or test suite of Meta's for the model was found. github.com/meta-models answered 403 to our fetcher, so its repositories are unchecked (0 of 25). On 4 November the BF16 repository's discussions page lists 13 open threads from 15 October to 1 November, 10 discussions and 3 pull requests (one a draft). A script counted 29 replies across the seven threads with replies, six discussions and the draft pull request, none from the repository owner or a Meta member, and their text wasn't read, so crash reports weren't separated from other posts (15 of 25, held from 15 October and flagged). No changelog, version tags or revision notes, and the commit history needs gated access (2 of 15). Released as a final checkpoint with no preview label from Meta, and both repositories unchanged since 15 October, though runtime support ships as launch-specific images and a branch, and Groq lists it as a preview (12 of 15). 13 + 0 + 15 + 2 + 12 = 42.\n- SWE-2 Flash (50): Model reading, for the surface it runs on, Devin Desktop and Devin CLI. Status page at www.devinstatus.com (Atlassian Statuspage) with Desktop Agent and Desktop Tab components besides the cloud ones, read on 4 November 2026 (20). The history feed for the 90 days from 6 August lists 'Devin unavailable' for 20 minutes on 13 August with no component named, read as across the board and a major as on the devin listing; Devin Cloud unavailable on 13 August and the web app down on 24 September name cloud surfaces this model doesn't use; the rest are degraded-performance incidents (new sessions on 3 September, 'multiple services' on 17 September, Ask Devin on 26 October, automation triggers on 4 November) and session queuing on 10 and 11 August, read as minor and a scheduled Devin Desktop extension marketplace upgrade on 7 August, read as maintenance. One major (10). No rate limit figures for the model or the CLI; usage is a daily and weekly quota with no numbers (0). No 429, backoff or retry guidance for this surface (0). Cognition publishes an SLA (25 February 2026) with uptime and response times that applies when included in an order form, read as a contractual SLA on Enterprise, as on other listings (10). The launch post makes it available today with no preview or beta label (10). 20 + 10 + 0 + 0 + 10 + 10 = 50.\n\n## Score by category\n\n| Category | Weight | Muse Spark Open | SWE-2 Flash | Edge |\n| --- | --- | --- | --- | --- |\n| Reliability | 16% (20 this run) | 42 | 50 | SWE-2 Flash +8 |\n| Performance | 10%, pending | pending | pending | not scored in this run |\n| Schema \u0026 documentation | 13% (16.2 this run) | 61 | 29 | Muse Spark Open +32 |\n| Agent ergonomics | 13% (16.2 this run) | 69 | 10 | Muse Spark Open +59 |\n| Security \u0026 auth | 14% (17.5 this run) | 44 | 53 | SWE-2 Flash +9 |\n| Payments \u0026 pricing | 10% (12.5 this run) | 50 | 20 | Muse Spark Open +30 |\n| Task success | 10%, pending | pending | pending | not scored in this run |\n| Maintenance \u0026 community | 7% (8.8 this run) | 60 | 53 | Muse Spark Open +7 |\n| Transparency \u0026 trust | 7% (8.8 this run) | 53 | 64 | SWE-2 Flash +11 |\n| Negative events | ≤15 | 0 | -2 | |\n| **Total** | | **53.4 · D** | **36.4 · F** | |\n\n## Facts side by side\n\n| Fact | Muse Spark Open | SWE-2 Flash |\n| --- | --- | --- |\n| Kind | Model API | Model API |\n| Vendor | Meta | Cognition AI, Inc. |\n| Hosted endpoint | no (local only) | no (local only) |\n| Transports | HTTP | |\n| Auth | None | OAuth or key |\n| Pricing | Free | Freemium |\n| x402 | no | no |\n| Licence | Muse Model Licence (source-available weights, not an OSI licence), free of charge with an attribution notice, a separate agreement with Meta before Model as a Service by a licensee group above $20M gross revenue, and the Muse Spark Open Usage Policy | Proprietary model, used through Devin Desktop and Devin CLI under Cognition's Platform Terms of Service. The weights are not public; Cognition says it is building a programme to share them with industry partners and researchers |\n| Read-only variant documented | no | no |\n| llms.txt | yes | yes |\n| Last release | 2026-10-15 | 2026-11-04 |\n| Terms last updated | 2026-10-15 | 2026-06-30 |\n| Privacy policy last updated | no document linked | 2026-03-09 |\n| Customer content may train models | not found in the text | yes, with an opt-out |\n| Terms restrict automated access | not found in the text | not found in the text |\n| Terms restrict benchmarking | not found in the text | yes |\n| Terms or service can change without notice | not found in the text | not found in the text |\n| Arbitration or class-action waiver | not found in the text | yes |\n\n## How the scores have moved\n\n### Muse Spark Open\n\n- -0.1 on 30 October 2026: checklist\n- no change on 28 October 2026: new evidence\n- +0.1 on 27 October 2026: checklist\n\n### SWE-2 Flash\n\nNo score change since it was listed.\n\n## Verdicts\n\n**Muse Spark Open.** Muse Spark Open runs on the owner's hardware in seven runtimes, with OpenAI-compatible servers in vLLM and SGLang, standard tool calls, a 1M-token context and Meta's own serving docs. Serving the 537 GB NVFP4 checkpoint takes four B300 GPUs, Meta's download is gated by manual review, and the licence adds attribution and a revenue threshold for model resellers.\n\n**SWE-2 Flash.** A new small coding model from Cognition that anyone can try free in Devin Desktop and Devin CLI until 30 November 2026, with strong vendor-reported coding scores. It has no model API, no published per-token price, context length or model card, and no public weights, so an agent can reach it only by driving Devin's own CLI.\n\n## Before you call either\n\n### Muse Spark Open\n\n1. Start vLLM or SGLang with `--reasoning-parser muse --tool-call-parser muse` (vLLM also `--enable-auto-tool-choice`), or tool calls come back as raw text\n2. Set `reasoning_effort` to `low` or `medium` for short tasks. The default is `xhigh`, reasoning can't be turned off, and other values return an error\n3. Stop only on `\u003c|eot|\u003e` and end of text. Stopping on `\u003c|eom|\u003e` cuts off tool calls and final answers\n4. Stream long generations and leave `max_tokens` room for reasoning, or `content` can come back empty with `finish_reason` `length`\n5. Don't call Meta Model API for this model. Use a self-hosted server or a hosted provider's model string, such as `meta-models/muse-spark-open`\n\n### SWE-2 Flash\n\n1. There is no endpoint to call. Run Devin CLI single-turn with `devin -p` and pick the model with `--model`; the SWE-2 Flash model ID isn't documented yet, so check `/model` in the CLI first\n2. Don't rely on the short name `swe`. The docs say it resolves to the latest model in the family, which may be SWE-2 rather than SWE-2 Flash\n3. Plan for billing to start on 1 December 2026. The free period covers Devin Desktop and CLI only, not Devin cloud sessions\n4. Treat Cognition's benchmark figures as the vendor's. No independent result for SWE-2 Flash was found\n5. Keep `credentials.toml` private. The token in it doesn't expire by default and carries the signed-in user's access\n\n## Questions\n\n### Which is better for AI agents, Muse Spark Open or SWE-2 Flash?\n\nMuse Spark Open scores 53.4 (D) on agent readiness against SWE-2 Flash's 36.4 (F), and leads in 4 of 7 scored categories. SWE-2 Flash leads on reliability, security \u0026 auth and transparency \u0026 trust.\n\n### Do Muse Spark Open and SWE-2 Flash need an API key?\n\nMuse Spark Open needs no key. SWE-2 Flash takes an API key or an OAuth sign-in.\n\n### Can an agent call Muse Spark Open and SWE-2 Flash without installing anything?\n\nNo hosted endpoint is listed for Muse Spark Open. No hosted endpoint is listed for SWE-2 Flash.\n\n\n## For agents\n\n- This comparison as JSON: https://www.anchorterminal.com/compare/muse-spark-open-vs-swe-2-flash.json, and with the fewest tokens: https://www.anchorterminal.com/compare/muse-spark-open-vs-swe-2-flash.min.md\n- Over MCP at https://www.anchorterminal.com/mcp (no key): `compare_tools {\"a\": \"muse-spark-open\", \"b\": \"swe-2-flash\"}`. From a terminal: `anchor compare muse-spark-open swe-2-flash`\n- Each listing in full: https://www.anchorterminal.com/api/v1/tools/muse-spark-open.json and https://www.anchorterminal.com/api/v1/tools/swe-2-flash.json\n\n## Other comparisons with Muse Spark Open or SWE-2 Flash\n\n- [Claude API vs Muse Spark Open](https://www.anchorterminal.com/compare/anthropic-api-vs-muse-spark-open.md)\n- [Claude API vs SWE-2 Flash](https://www.anchorterminal.com/compare/anthropic-api-vs-swe-2-flash.md)\n- [Antseed vs Muse Spark Open](https://www.anchorterminal.com/compare/antseed-vs-muse-spark-open.md)\n- [Antseed vs SWE-2 Flash](https://www.anchorterminal.com/compare/antseed-vs-swe-2-flash.md)\n- [BlockRun.AI vs Muse Spark Open](https://www.anchorterminal.com/compare/blockrun-ai-vs-muse-spark-open.md)\n- [BlockRun.AI vs SWE-2 Flash](https://www.anchorterminal.com/compare/blockrun-ai-vs-swe-2-flash.md)\n- [Cerebras Inference vs Muse Spark Open](https://www.anchorterminal.com/compare/cerebras-inference-vs-muse-spark-open.md)\n- [Cerebras Inference vs SWE-2 Flash](https://www.anchorterminal.com/compare/cerebras-inference-vs-swe-2-flash.md)\n- [Cloudflare AI Gateway vs Muse Spark Open](https://www.anchorterminal.com/compare/cloudflare-ai-gateway-vs-muse-spark-open.md)\n- [Cloudflare AI Gateway vs SWE-2 Flash](https://www.anchorterminal.com/compare/cloudflare-ai-gateway-vs-swe-2-flash.md)\n- [Cloudflare Workers AI vs Muse Spark Open](https://www.anchorterminal.com/compare/cloudflare-workers-ai-vs-muse-spark-open.md)\n- [Cloudflare Workers AI vs SWE-2 Flash](https://www.anchorterminal.com/compare/cloudflare-workers-ai-vs-swe-2-flash.md)\n- [Cohere Chat API (Command models) vs Muse Spark Open](https://www.anchorterminal.com/compare/cohere-chat-vs-muse-spark-open.md)\n- [Cohere Chat API (Command models) vs SWE-2 Flash](https://www.anchorterminal.com/compare/cohere-chat-vs-swe-2-flash.md)\n- [DeepInfra vs Muse Spark Open](https://www.anchorterminal.com/compare/deepinfra-vs-muse-spark-open.md)\n- [DeepInfra vs SWE-2 Flash](https://www.anchorterminal.com/compare/deepinfra-vs-swe-2-flash.md)\n- [DeepSeek API vs Muse Spark Open](https://www.anchorterminal.com/compare/deepseek-api-vs-muse-spark-open.md)\n- [DeepSeek API vs SWE-2 Flash](https://www.anchorterminal.com/compare/deepseek-api-vs-swe-2-flash.md)\n- [Glimpse vs Muse Spark Open](https://www.anchorterminal.com/compare/elorian-glimpse-vs-muse-spark-open.md)\n- [Glimpse vs SWE-2 Flash](https://www.anchorterminal.com/compare/elorian-glimpse-vs-swe-2-flash.md)\n- [Fireworks AI Inference vs Muse Spark Open](https://www.anchorterminal.com/compare/fireworks-inference-vs-muse-spark-open.md)\n- [Fireworks AI Inference vs SWE-2 Flash](https://www.anchorterminal.com/compare/fireworks-inference-vs-swe-2-flash.md)\n- [Gemini Developer API vs Muse Spark Open](https://www.anchorterminal.com/compare/gemini-api-vs-muse-spark-open.md)\n- [Gemini Developer API vs SWE-2 Flash](https://www.anchorterminal.com/compare/gemini-api-vs-swe-2-flash.md)\n- [GroqCloud vs Muse Spark Open](https://www.anchorterminal.com/compare/groq-vs-muse-spark-open.md)\n- [GroqCloud vs SWE-2 Flash](https://www.anchorterminal.com/compare/groq-vs-swe-2-flash.md)\n- [Jais 3 vs Muse Spark Open](https://www.anchorterminal.com/compare/jais-3-vs-muse-spark-open.md)\n- [Jais 3 vs SWE-2 Flash](https://www.anchorterminal.com/compare/jais-3-vs-swe-2-flash.md)\n- [Meta Model API vs Muse Spark Open](https://www.anchorterminal.com/compare/meta-model-api-vs-muse-spark-open.md)\n- [Meta Model API vs SWE-2 Flash](https://www.anchorterminal.com/compare/meta-model-api-vs-swe-2-flash.md)\n- [MiniMax API vs Muse Spark Open](https://www.anchorterminal.com/compare/minimax-api-vs-muse-spark-open.md)\n- [MiniMax API vs SWE-2 Flash](https://www.anchorterminal.com/compare/minimax-api-vs-swe-2-flash.md)\n- [Mistral AI API vs Muse Spark Open](https://www.anchorterminal.com/compare/mistral-api-vs-muse-spark-open.md)\n- [Mistral AI API vs SWE-2 Flash](https://www.anchorterminal.com/compare/mistral-api-vs-swe-2-flash.md)\n- [Muse Spark Open vs Nemotron 3.5 Super VL](https://www.anchorterminal.com/compare/muse-spark-open-vs-nemotron-3-5-super-vl.md)\n- [Muse Spark Open vs Novita AI](https://www.anchorterminal.com/compare/muse-spark-open-vs-novita-ai.md)\n- [Muse Spark Open vs OpenAI API](https://www.anchorterminal.com/compare/muse-spark-open-vs-openai-api.md)\n- [Muse Spark Open vs OpenRouter](https://www.anchorterminal.com/compare/muse-spark-open-vs-openrouter.md)\n- [Muse Spark Open vs Prism Inference](https://www.anchorterminal.com/compare/muse-spark-open-vs-prism-inference.md)\n- [Muse Spark Open vs Router.com (Ramp Router)](https://www.anchorterminal.com/compare/muse-spark-open-vs-ramp-router.md)\n- [Muse Spark Open vs Beam 1.0](https://www.anchorterminal.com/compare/muse-spark-open-vs-reflection-beam.md)\n- [Muse Spark Open vs RouterPlus](https://www.anchorterminal.com/compare/muse-spark-open-vs-routerplus.md)\n- [Muse Spark Open vs SambaCloud](https://www.anchorterminal.com/compare/muse-spark-open-vs-sambanova.md)\n- [Muse Spark Open vs SiliconFlow](https://www.anchorterminal.com/compare/muse-spark-open-vs-siliconflow.md)\n- [Muse Spark Open vs Simplismart](https://www.anchorterminal.com/compare/muse-spark-open-vs-simplismart.md)\n- [Muse Spark Open vs StepFun API](https://www.anchorterminal.com/compare/muse-spark-open-vs-stepfun-api.md)\n- [Muse Spark Open vs Tencent Cloud TokenHub](https://www.anchorterminal.com/compare/muse-spark-open-vs-tencent-tokenhub.md)\n- [Muse Spark Open vs Together AI Inference](https://www.anchorterminal.com/compare/muse-spark-open-vs-together-inference.md)\n- [Muse Spark Open vs Vercel AI Gateway](https://www.anchorterminal.com/compare/muse-spark-open-vs-vercel-ai-gateway.md)\n- [Nemotron 3.5 Super VL vs SWE-2 Flash](https://www.anchorterminal.com/compare/nemotron-3-5-super-vl-vs-swe-2-flash.md)\n- [Novita AI vs SWE-2 Flash](https://www.anchorterminal.com/compare/novita-ai-vs-swe-2-flash.md)\n- [OpenAI API vs SWE-2 Flash](https://www.anchorterminal.com/compare/openai-api-vs-swe-2-flash.md)\n- [OpenRouter vs SWE-2 Flash](https://www.anchorterminal.com/compare/openrouter-vs-swe-2-flash.md)\n- [Prism Inference vs SWE-2 Flash](https://www.anchorterminal.com/compare/prism-inference-vs-swe-2-flash.md)\n- [Router.com (Ramp Router) vs SWE-2 Flash](https://www.anchorterminal.com/compare/ramp-router-vs-swe-2-flash.md)\n- [Beam 1.0 vs SWE-2 Flash](https://www.anchorterminal.com/compare/reflection-beam-vs-swe-2-flash.md)\n- [RouterPlus vs SWE-2 Flash](https://www.anchorterminal.com/compare/routerplus-vs-swe-2-flash.md)\n- [SambaCloud vs SWE-2 Flash](https://www.anchorterminal.com/compare/sambanova-vs-swe-2-flash.md)\n- [SiliconFlow vs SWE-2 Flash](https://www.anchorterminal.com/compare/siliconflow-vs-swe-2-flash.md)\n- [Simplismart vs SWE-2 Flash](https://www.anchorterminal.com/compare/simplismart-vs-swe-2-flash.md)\n- [StepFun API vs SWE-2 Flash](https://www.anchorterminal.com/compare/stepfun-api-vs-swe-2-flash.md)\n- [SWE-2 Flash vs Tencent Cloud TokenHub](https://www.anchorterminal.com/compare/swe-2-flash-vs-tencent-tokenhub.md)\n- [SWE-2 Flash vs Together AI Inference](https://www.anchorterminal.com/compare/swe-2-flash-vs-together-inference.md)\n- [SWE-2 Flash vs Vercel AI Gateway](https://www.anchorterminal.com/compare/swe-2-flash-vs-vercel-ai-gateway.md)\n", "meta": { "attribution": "Anchor Terminal (https://www.anchorterminal.com)", "attributionRequired": true, "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)", "docs": "https://www.anchorterminal.com/docs/", "generatedAt": "2026-11-05", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.7", "openapi": "https://www.anchorterminal.com/openapi.json", "preview": false, "run": "2026-10-01", "runLabel": "October 2026 research run", "terms": "https://www.anchorterminal.com/terms/#data" }, "page": { "breadcrumbs": [ { "name": "Home", "url": "https://www.anchorterminal.com/" }, { "name": "Compare", "url": "https://www.anchorterminal.com/compare/" }, { "name": "Muse Spark Open vs SWE-2 Flash", "url": "" } ], "description": "Muse Spark Open scores 53.4 (D) to SWE-2 Flash's 36.4 (F) for llm inference. Prices, MCP, x402, uptime and agent notes side by side.", "facts": [ "Muse Spark Open D 53.4", "SWE-2 Flash F 36.4", "scores" ], "h1": "Muse Spark Open vs SWE-2 Flash", "image": "https://www.anchorterminal.com/assets/og/default.png", "path": "/compare/muse-spark-open-vs-swe-2-flash", "published": "2026-10-01", "section": "tools", "title": "Muse Spark Open vs SWE-2 Flash for AI agents (2026) | Anchor Terminal", "toc": null, "updated": "2026-11-04", "url": "https://www.anchorterminal.com/compare/muse-spark-open-vs-swe-2-flash" }, "tokens": { "markdown": 7000, "slim": 3880 }, "version": 1 }