Skip to main content
AI Choice Engine

Meta

Muse Spark Open 816B A42B

Muse Spark Open 816B A42B is a Meta model with 1.05M context, Not verified in / Not verified out per million tokens, last verified October 25, 2026. Open weights: yes. Public-board scores are linked to the publisher; vendor-only figures stay in the watch-out, not on the board.

Meta’s 816B/42B multimodal MoE checkpoint for self-hosted agents and coding, with its own licence and native reasoning controls.

Recorded profile verification October 25, 2026Individual provider fields may change

AI model specification details
SpecificationMuse Spark Open 816B A42B
ProviderMeta
TierFrontier
Context window1.05M
Max outputNot verifiedUnverified
Input / 1M tokensNot verifiedUnverified
Output / 1M tokensNot verifiedUnverified
WeightsOpen
Parameters816B total / 42B active (MoE)
Reasoning levelslow, medium, high, xhigh
Modalitiestext, image, video, audio
LicenseMuse Model License (Spark Open)
ReleasedNot verifiedUnverified

Pricing tiers: Downloadable weights for self-hosting or separately priced providers; Meta’s hosted Spark API rates do not price this checkpoint. Published weight footprints are about1.65 TB BF16 or537 GB NVFP4, excluding KV cache, activations and runtime overhead; those are not per-token prices or universal hardware-fit guarantees. First-party hosted token prices remain unknown.

Best for

  • Large-scale self-hosted agents
  • Evaluating exact native checkpoints
  • Provider-specific open-weight deployments

Watch out

Native reasoning is always on: low/medium/high/xhigh, default xhigh, mapped to strengths 8/32/256/512; none and max are rejected. A numeric Reasoning strength system line takes precedence. Parallel tool calls are supported. Licence requires Built with Muse Spark Open attribution including derivatives created using outputs, notices and Usage Policy compliance. MaaS operators whose licensee/affiliate aggregate gross revenue exceeds US$20M in any consecutive 12 months need notification and a separate Meta agreement; internal/research/non-MaaS and embedded own-product features outside MaaS are excepted. Do not inherit hosted Spark’s controls, data-use contract, exact capabilities or scores.

Common questions

Muse Spark Open 816B A42B

Answered from the verified figures on this page rather than general guidance.

What is Muse Spark Open 816B A42B's context window?
Muse Spark Open 816B A42B accepts about 1.05M tokens of context. That only matters if you routinely send very long documents, large codebases, or multi-turn histories that approach that limit.
What is Muse Spark Open 816B A42B best for?
Muse Spark Open 816B A42B is a frontier tier from Meta. It suits large-scale self-hosted agents, evaluating exact native checkpoints, provider-specific open-weight deployments. Native reasoning is always on: low/medium/high/xhigh, default xhigh, mapped to strengths 8/32/256/512; none and max are rejected. A numeric Reasoning strength system line takes precedence. Parallel tool calls are supported. Licence requires Built with Muse Spark Open attribution including derivatives created using outputs, notices and Usage Policy compliance. MaaS operators whose licensee/affiliate aggregate gross revenue exceeds US$20M in any consecutive 12 months need notification and a separate Meta agreement; internal/research/non-MaaS and embedded own-product features outside MaaS are excepted. Do not inherit hosted Spark’s controls, data-use contract, exact capabilities or scores.
Can I self-host Muse Spark Open 816B A42B?
Muse Spark Open 816B A42B publishes open weights, but self-hosting depends on the licence, hardware footprint, quantisation quality, and serving stack. A hosted API is often cheaper until you have measured throughput and concurrency on your own hardware.