Muse Spark Open vs xAI API (Grok)

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 <tool_output>, 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.

xAI API (Grok) 97

Model reading. Team API keys with ACLs per endpoint and per model, QPM and TPM caps, expiry, and create, rotate, disable and delete through the Management API, plus documented mTLS. GitHub secret scanning disables leaked keys and emails the owner (30). The security FAQ says SpaceXSI never trains on API inputs or outputs without explicit permission (20). Requests and responses are kept 30 days for abuse auditing, then deleted, and team admins can switch on zero data retention themselves, confirmed by an x-zero-data-retention header on every response (15). Console audit log and a Management API audit-events endpoint, plus optional per-key request logging (15). security.txt valid to 2027-06-29 with a HackerOne bug bounty and a separate safety contact for prompt-injection issues. SOC 2 Type 2 is claimed in the security FAQ with the reports behind an NDA in the trust centre, so 17 of 20. Remote MCP doesn't support require_approval, which the model checklist doesn't score but reviewers should see. New in October, a guidelines array lets the caller make the model decline nsfw, violence or profanity content, which the checklist doesn't score.

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.

xAI API (Grok) 20

No x402, MPP or L402 (0). Per-token and per-call prices published without a login on docs.x.ai/developers/pricing (20). No free tier found for the text API, and the quickstart says to load the account with credits (0). Sign-up is in the browser at console.x.ai. The Management API can create keys and top up prepaid credit, but only with a management key made by a person in the console, so no line credit (0).

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.

xAI API (Grok) 88

OpenAPI 3.1 document at docs.x.ai/openapi.json, 39 paths and 205 schemas on 6 November, covering Responses, Chat Completions, files, images, videos and models. The batch endpoints documented in the REST reference still aren't in it (20 of 25, applied as for gemini-api's missing Interactions API). llms.txt and a Markdown twin of every page, plus a docs MCP server (10). The models page says which model to use for code, chat, images, video and voice, and reference pages list each parameter. Reference not read in full (18 of 20). Structured outputs are guaranteed to match the schema for supported keywords, and tool arguments always conform with strict implicitly true (15). Examples in Python, JavaScript and curl on every guide. The debugging page lists status codes with fixes, but the cause cells for 400, 403, 404 and 415 are still empty (13 of 15). Release notes are grouped by month without day dates, and model slugs come as aliases and dated versions (12 of 15).

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.

xAI API (Grok) 90

Model reading of the checklist (tool use, structured output, caching, context, batch, SDKs, errors). Function calling with parallel calls on by default and tool_choice of auto, required, none or a named function (20). Guaranteed structured outputs (15). Automatic prompt caching, cached input at 0.25x on grok-4.7 ($0.50 per million) and 0.16x on grok-4.3 (15). 500,000-token context on grok-4.7 and 1,000,000 on grok-4.3, 13 of 15 for the default model's 500K. The Batch API is now 20% off on grok-4.7, 4.6 and 4.5 as well as the older models, with batch requests outside the rate limits (10 of 10, up from 5 on 6 October). One first-party SDK, xai-sdk for Python, with the OpenAI SDKs and Vercel's AI SDK documented as alternatives (5 of 10). Status codes documented with fixes, no Retry-After header found (12 of 15).

Muse Spark Open 60

Read for an open-weight model released three weeks ago. Released on 15 October 2026 (30). One release, BF16 and NVFP4 on the same day, with no history of revisions, successors or notice periods, and Meta Model API has no deprecation policy either (6 of 20). None of the 29 replies counted by script on 4 November across the BF16 repository's discussions came from Meta. Feedback goes to the discussions or museoss@meta.com, and Meta's docs and developer blog carry dated posts (8 of 25). Day-one support in vLLM, SGLang, TensorRT-LLM, Transformers, llama.cpp, MLX and Unsloth with Meta's own recipes, though SGLang's is a dev image and TensorRT-LLM's a feature branch (12 of 15). Current dependencies (Transformers 5.19), but no CI visible to us (4 of 10). 30 + 6 + 8 + 12 + 4 = 60.

xAI API (Grok) 77

Model reading. grok-imagine-image-quality and grok-imagine-image-pro retired on 2 November 2026 as announced in September, Grok Voice 3.0 and custom moderation controls in October, Grok 4.7 on 21 September (30). No minimum notice period found. Retirements come as dated migration guides, retired slugs redirect to a replacement rather than failing, billed at the replacement's price, and grok-voice-2.0 was deprecated in October with no end date (4 of 12). Shutdowns in the last 90 days with dates given in advance, grok-voice-transcribe-1.0 on 2 October and grok-imagine-image-quality on 2 November (8 of 8). Release notes several times a month but dated only by month, support by email and a Discord #help channel, not tested. On xai-sdk-python a collaborator commented on 2 of the 16 most recent threads with replies, and most outside pull requests since 23 October are open without a maintainer reply (10 of 15, down from 12). The repository has CI, release and version-bump workflows (5 of 10). One current first-party SDK, xai-sdk 1.21.0 tagged on 23 October, Python only (10 of 15). Supported runtimes stated, Python 3.10 or later (10).

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.

xAI API (Grok) 55

Status page at status.x.ai with components for the global and US API endpoints, voice, the consumer apps and, since the Cursor merger, Cursor's services, with an RSS feed (20). The feed, read on 6 November, now holds 267 incidents back to 11 August, so the whole 90-day window is readable. On the API components there were high error rates on grok-4.6 and grok-4.5 from 21:57 to 23:28 UTC on 17 August (about 90 minutes) and a models outage on the global and US endpoints from 13:30 to 17:09 UTC on 3 September (about 3 hours 40 minutes), so several majors. A 26-minute datacenter outage across the API on 22 September, voice disruptions of 12 and 28 minutes on 10 September and 6 October, a 63-minute voice and Imagine degradation on 30 October and a 37-second model degradation on 24 October add to the record (0). Rate limits published per model and tier, grok-4.7 at Tier 0 is 150 requests a second and 50M TPM, with the per-second cap derived as RPM divided by 48 (15). The rate-limits page shows exponential backoff on 429, but no Retry-After header or overload guidance was found (10 of 15). No SLA found in the docs or llms.txt (0). The Responses API has been GA since September 2025 and grok-4.7 carries no preview label (10).