Muse Spark 1.1

// Overview

Muse Spark 1.1 is Meta Superintelligence Labs' upgrade to the original Muse Spark, released on July 9, 2026. Meta positions it as a multimodal reasoning model built for agentic work — long-horizon coding, computer use, tool orchestration, and multimodal understanding — with major gains over the April 2026 Muse Spark on all four axes.

The model ships with a 1 million-token context window, adds multi-agent orchestration and computer use, and demonstrates zero-shot generalization to new native tools, MCP servers, and custom skills. It runs in Thinking mode inside the Meta AI app and on meta.ai for consumers, and opens the Meta Model API in public preview for developers — the first time Meta has charged for access to its own models.

API pricing is $1.25 per million input tokens and $4.25 per million output tokens, placing Muse Spark 1.1 in the same tier as Claude Haiku 4.5 and GPT-5.6 Luna. Meta's launch table shows Muse Spark 1.1 leading peers on tool-use and tool-augmented reasoning while trailing Claude Opus 4.8 and GPT-5.6 on coding and multimodal benchmarks.

// Benchmarks

This model's scores

1. MCP Atlas (tool use)88.1%
2. JobBench (professional tool use)54.7%
3. Humanity's Last Exam (with tools)62.1%
4. OSWorld-Verified (computer use)80.8%
5. SWE-Bench Pro (coding)61.5%
6. DeepSWE 1.1 (long-horizon coding)53.3%
7. BabyVision (visual reasoning)76.3%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Meta Model API public preview pricing for US developers; the model is also available free in Thinking mode inside the Meta AI app and on meta.ai.

Pricing source ↗

// Strengths

• Leads Meta's published launch table on tool use (MCP Atlas 88.1), professional tool use (JobBench 54.7), and tool-augmented reasoning (Humanity's Last Exam with tools 62.1)
• 1M-token context window with active context management — remembers earlier actions, retrieves earlier work, and compacts to preserve critical steps
• Multi-agent orchestration for long-horizon agentic tasks
• Zero-shot generalization to new native tools, MCP servers, and custom skills
• Priced at $1.25/$4.25 per million tokens — roughly a quarter of comparable Anthropic and OpenAI rates at similar tiers

// Best for

• Reach for it when you need a long-context agent that can drive tools, browsers, and computers across multi-step workflows.
• Reach for it for large code migrations and bug fixes across big codebases where the 1M-token window and context compaction matter.
• Reach for it when you want a lower-cost frontier-tier agentic model on the Meta Model API to prototype workflows against Claude Haiku 4.5 or GPT-5.6 Luna.

// How to access

// Muse Spark — every version

The full lineage of the Muse Spark line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.