Muse Spark Open vs Vertex AI (Gemini Enterprise Agent Platform)

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.

Vertex AI (Gemini Enterprise Agent Platform) 99

Google Cloud status dashboard with per-product history (20). Its incident feed, read on 6 November, has no incident naming Vertex Gemini API, Gemini or the Agent Platform in the 90 days. The regional incidents of 20 August, 1 September and 20 October (europe-west2) and the Cloud Storage incident of 8 and 9 October didn't list it among affected products (30). Standard PayGo publishes usage tiers by 30-day spend with baseline throughput, 2,000,000 to 50,000,000 tokens a minute for Flash models and 500,000 to 10,000,000 for Pro, plus quota tables for other methods (15). A 429 page says to use the global endpoint and truncated exponential backoff, and a retry-strategy page (15). SLA of 99.5 per cent a month on generateContent and streamGenerateContent, 95 per cent for short-term availability models, and a latency SLO on Provisioned Throughput. The 429 page says pay-as-you-go 429s don't count toward the SLA error rate (10). Gemini 4 Argon reached GA on 13 October and Gemini 3.8 Flash moved to the 12-month tier, and the v1 API, 3.5 Flash-Lite and Claude Opus 5.5 are GA, but express mode was Preview at the 6 October check (9 of 10, up from 8). Up from 98 on 6 October.

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.

Vertex AI (Gemini Enterprise Agent Platform) 97

Model reading. OAuth 2.0 with IAM roles and service accounts, which the docs recommend for production, with API keys for testing. No key in a query string found in the docs we read (30). The training restriction in the Service Specific Terms says Google won't use customer data to train or fine-tune any model without permission, GA and pre-GA alike (20). Default in-memory caching for 24 hours that can be disabled, flagged prompts logged up to 90 days unless an abuse-monitoring exception is granted or the customer has a Cloud Master Agreement, the Interactions API storing by default, and Claude Fable 5 and Mythos 5 requiring 30-day retention shared with Anthropic (12 of 15, a judgement call matching the Fable exception on the Anthropic and Bedrock listings). Cloud Audit Logs, with Data Access logs off until enabled, and optional request-response logging to BigQuery (15). google.com security.txt valid to 2030-04-01, the Google VRP, and Generative AI on the platform in ISO 27001, 27017 and 27018 and SOC 1, 2 and 3 scope per the 1 October services-in-scope check for vertex-ai-tuning, not re-read on 6 October (20).

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.

Vertex AI (Gemini Enterprise Agent Platform) 90

Closed service under the Google Cloud terms, with the Gen AI SDKs under Apache-2.0 and a public discovery document (18, as for Bedrock's open service models). The zero data retention and abuse monitoring pages agree with the Cloud terms and give periods for each case, 24 hours in memory, up to 90 days for flagged prompts, 30 days for Advanced AI models, 3 days for Search grounding logs and 30 days for Maps grounding, and say which can't be switched off (27 of 30). A dated model versions table with release and retirement dates and a stated 12-month or 45-day policy, but on 6 November it still listed Gemini 2.5 Pro, Flash and Flash-Lite as available with a 20 October 2026 retirement date and left them out of the retired table (18 of 20, down from 20). The Cloud subprocessor list with locations, per the 1 October vertex-ai-tuning check, regional and multi-region endpoints, and per-model ML processing locations on the partner model cards (20).

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.

Vertex AI (Gemini Enterprise Agent Platform) 94

Model reading. Keyless discovery document for the Agent Platform API v1, revision 20260930, with 1,128 methods and 1,493 schemas (25). No llms.txt, but every docs page answers as Markdown at .md.txt, counted as for vertex-ai-tuning (10). Model cards give limits, capabilities and retirement dates, and the 3.8 Flash card says when to use 3.7 Flash instead, but the docs mix the Vertex AI and Agent Platform names, and the OpenAI page tells new users to call the Gemini Developer API instead (16 of 20). Structured output with response_schema and response_mime_type: application/json always follows the schema per the docs, and complex schemas fail with a 400 (15). Examples on every page and a 429 guide, but we found no full error code list for generateContent (13 of 15). Versioned v1 and v1beta1 APIs and release notes with entries on most working days, newest 5 November 2026 (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.

Vertex AI (Gemini Enterprise Agent Platform) 91

Model reading. Gemini Omni 1.2 Flash GA on 5 November, Grok 4.7 GA on 27 October, Gemini 4 Argon GA on 13 October and Claude Haiku 5.5 on 7 October 2026 (30). Stable models stay at least 12 months, and since 6 October Gemini 4 Argon and Gemini 3.8 Flash are in that tier, while 3.7 and 3.6 Flash remain short-term with at least 45 days' notice once a date is set (10 of 12, up from 8). Gemini 2.5 Pro, Flash and Flash-Lite were due to retire on 20 October, three models on one date after more than 12 months, with replacements named. We didn't check partner model retirements (6 of 8, down from 8). Dated release notes on most working days, newest 5 November, and Cloud Customer Care (15). python-genai replies not sampled this run (5 of 10). Current SDKs, google-genai 3.0.0 on 6 November and @google/genai 3.0.0, and google-cloud-aiplatform 2.7.0 on 28 October 2026 (15). Supported runtimes stated, Python 3.10 or later and Node 20 or later, per the 6 October check (10). Unchanged at 91.

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.

Vertex AI (Gemini Enterprise Agent Platform) 97

Model reading of the checklist (tool use, structured output, caching, context, batch, SDKs, errors). Function calling with FunctionCallingConfig mode ANY and allowedFunctionNames to force a named tool (20). Schema-guaranteed JSON output (15). Implicit and explicit context caching, cached input at 0.1x, $0.075 per million on 3.8 Flash through 31 December (15). 1,048,576-token context on 3.8 Flash, a unified 1,048,576-token input and output limit on Gemini 4 Argon, and 1,000,000 on Claude Opus 5.5 (15). Flex/Batch at half price on Gemini (10). Gen AI SDKs for Python and JavaScript among others, and the OpenAI libraries work with Google Cloud auth (10). 429 handling documented per quota framework, with standard Google status codes elsewhere (12 of 15).