Hugging Face Inference Providers vs Muse Spark Open

Hugging Face Inference Providers 79

Model reading. Fine-grained user access tokens that can be limited to the Make calls to Inference Providers permission, rotated and revoked, plus a public revocation endpoint for leaked tokens, and organisation approval of fine-grained tokens on paid plans (30). Hugging Face says it doesn't store user data for training, but data handling past the router follows each provider's policy, with no per-request filter to exclude providers that train or retain (12 of 20). Request bodies and responses aren't stored when routed and debug logs are kept up to 30 days without user data or tokens, with provider retention outside that statement (10 of 15). Usage by model and provider in account settings, an Enterprise usage API, and Team and Enterprise admins can disable providers and set a spending limit (13 of 15). SOC 2 Type 2 for the Hub, of which the docs say Inference Providers is a feature, and security.txt valid to 2030 per the Inference Endpoints check; no bug bounty found (14 of 20).

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.

Hugging Face Inference Providers 20

No x402, MPP or L402 (0). Per-token prices for every model and provider published without a login in the router's /v1/models and on model pages (20). No free tier since 7 October 2026, when free users lost their included monthly credits; PRO's $2.00 of credits needs a subscription (0). Sign-up, token creation and credit purchase are browser steps (0).

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.

Hugging Face Inference Providers 80

Model reading. huggingface_hub 2.4.0 tagged on 29 October 2026 and the Inference Providers docs updated daily by an automated job, last on 6 November (30). huggingface_hub 2.2.0, 2.3.0 and 2.4.0 and @huggingface/inference 4.13.28 to 4.13.32 in the 90 days (20). No dedicated changelog; the Hub has a forum and Discord, and GitHub issue replies were not read (8 of 15). Current official SDKs in Python and JavaScript (15). The Python client supports Python 3.10 and later, per the Inference Endpoints check; CI not read (7 of 10).

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.

Hugging Face Inference Providers 71

Closed service under Terms of Service that name Inference Providers, with an Apache-2.0 Python client and an MIT JavaScript client (20). The security page and the docs agree on no training, no stored bodies and 30-day logs, but the privacy policy dates from 28 March 2023 and each provider's own data policy governs the rest (18 of 30). No deprecation policy found. Provider support changes and the 7 October removal of free credits are dated only in the docs repository history (5 of 20). The privacy policy lists 11 subprocessors with countries per the Inference Endpoints check, and the docs name every provider a request can reach, though not where each runs (15 of 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.

Hugging Face Inference Providers 25

Hosted lines. status.huggingface.co is a Better Stack page with 90-day component history for the Hub, Inference Endpoints, Spaces, the CDNs and Jobs, but no component for Inference Providers or the router (10 of 20). With no component, there is no readable incident history for the surface agents call, and the feed's 23 items from 10 June to 7 October 2026 don't name it (5). No rate limits for the router were found; the Hub rate-limits page gives numbers for Hub APIs, resolvers and pages only (0). No 429 or overload guidance for the router was found (0). No SLA found (0). Chat completions on the router is generally available; the Responses API is labelled beta (10). Routing between providers with :fastest or :cheapest softens a single provider's outage, which this checklist doesn't score.

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.

Hugging Face Inference Providers 58

Router reading of the model checklist (tool use, structured output, caching, context, batch, SDKs, errors), as for openrouter. Tool calling through the OpenAI-compatible API with tool_choice, support flagged per provider in /v1/models (15 of 20). Structured output documented, support flagged per provider and false on some (10 of 15). No prompt caching guidance found, though the Enterprise usage API counts cached input tokens (3 of 15). Routes to models with up to 1,048,576 tokens of context per the model list, a judgement call since a router has no default model (15). No batch API (0 of 10). Official clients in Python and JavaScript, plus any OpenAI SDK (10). Errors not documented (5 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.