Muse Spark

Muse Spark is Meta's proprietary multimodal reasoning model for agentic and coding work, served on Meta Model API and in Muse Code, with a 1M-token context window and OpenAI/Anthropic SDK-compatible endpoints.

Related guides

Browse curated shortlists where this tool appears.

AI tools for

All use cases · All industries

Alternatives & similar tools

Other tools in our directory you may want to compare.

Also mentioned

Gemini (Google Flash / API workhorse models)
Claude (Anthropic API + Claude Code)
ChatGPT / OpenAI API (Codex and ChatGPT agents)
DeepSeek (cost-focused API models)
Grok (xAI API models)

Features & details

Full overview from our catalog (read-only reference).

Category (short)

Sponsored note

Target audience

Builders who want Meta's proprietary agentic/coding model with clear per-token pricing and Muse Code as a first-party harness. Not ideal for: teams that need open weights (use Muse Glimmer / Llama lines instead), or buyers who refuse any cloud API and Meta account.

Best for

Pricing notes

Verified September 2026 at https://dev.meta.ai/docs/pricing-rate-limits.md and https://dev.meta.ai/docs/models.md. Standard tier (muse-spark-1.3, 1.2, 1.1): $1.25 / 1M input, $4.25 / 1M output, $0.15 / 1M cached input; prompts not used for training. Contributor tier (muse-spark-1.3-contributor, 1.2-contributor): $0.10 / 1M input, $0.20 / 1M output, $0.002 / 1M cached input; Meta may train on prompts and completions. Web search grounding: $2.50 per 1,000 search queries plus tokens. Rate limits (per team): Standard 3,000 RPM / 4M TPM; Contributor 100 RPM / 3M TPM. Muse Spark 1.3 max reasoning was described as limited preview at the Sep 2, 2026 launch post; confirm live model ids in the API catalog. Distinct from Muse Image ($0.01/image), Muse Voice Transcribe ($0.18/hour audio), and open-weight Muse Glimmer.

Pros & cons

Editorial notes to help compare fit before opening the vendor site.

Pros

Clear Standard vs Contributor pricing on official docs
1M context with agent-oriented API primitives
First-party Muse Code harness plus SDK-compatible API
1.3 checkpoint focused on long-horizon agents and coding efficiency

Cons

Proprietary cloud model (no Spark open weights)
Contributor tier trades privacy for cheaper tokens
Audio understanding on 1.3 is limited per Meta docs
max reasoning access may still be gated for some accounts

Review notes

Context from the listing review and editorial research.

NEW tool. Product name for the directory is Muse Spark (model family). Current recommended API id: muse-spark-1.3 (Sep 2026). DISAMBIGUATION: not Llama open weights; not Muse Image / Muse Voice Transcribe / Muse Glimmer (sibling Meta Muse families). Muse Code is the first-party coding harness (`curl -fsSL https://dev.meta.ai/install.sh | bash` per Meta's 1.3 post). Related site content: reviews-muse-spark-1-3, compare-gemini-3-8-flash-vs-muse-spark-1-3. Artificial Analysis and Meta both describe strong agentic/coding positioning for 1.3; treat vendor and third-party benches as non-production until you run your own evals.

Extended features

In-depth description and capability notes.

Latest checkpoint Muse Spark 1.3 (muse-spark-1.3) recommended for new work; 1.2 and 1.1 still listed 1,048,576-token context window Multimodal input: text, image, video, PDF (audio understanding limited on 1.3; prefer 1.2 or Muse Voice Transcribe for speech) Agent primitives: parallel tool calls, streamed tool-call arguments, reasoning across turns Muse Code coding-agent harness plus Meta Model API (Responses, Chat Completions, Messages) OpenAI SDK, Anthropic SDK, and agent CLI drop-in base URL https://api.meta.ai/v1 Standard vs Contributor pricing tiers (Contributor allows Meta to train on your traffic) Optional web search grounding tool (separate query fee)

Use cases

Developers and teams run long-horizon coding agents, multi-step tool loops, and multimodal reasoning through Muse Code or any OpenAI-compatible client pointed at Meta Model API.