Meta has launched Muse Spark 1.1, a multimodal reasoning model for agentic AI tasks.
Alongside the model launch, Meta introduced the Meta Model API in public preview, marking the first time developers can build applications using Muse Spark models through the company’s own API platform.
The company said Muse Spark 1.1 is available in Thinking mode through the Meta AI app and Meta AI website. For API access, Meta has priced the model at $1.25 per million input tokens and $4.25 per million output tokens.
The release follows the launch of Muse Image earlier this week as Meta continues to build its AI offerings under Meta Superintelligence Labs.
“Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks, with major gains in tool and computer use, coding, and multimodal understanding,” the company said in a blog post.
Meta said the model is built to execute tasks across multiple applications and external services, with support for native tools, model context protocol (MCP) servers and custom skills. The company said the model can coordinate multiple AI agents by planning tasks, delegating work to sub-agents and combining their outputs to reduce execution time.
The model supports a context window of one million tokens, allowing it to retain information from earlier interactions, retrieve relevant context and compress conversations while preserving information required for later tasks.
According to Meta, Muse Spark 1.1 scored 88.1 on MCP Atlas, ahead of Anthropic’s Opus 4.8 (82.2) and OpenAI’s GPT-5.5 (xhigh) (75.3). On JobBench, it also led both models with a score of 54.7, compared with 48.4 for Opus 4.8 and 38.3 for GPT-5.5. On Humanity’s Last Exam with tools and Finance Agent v2, Muse Spark 1.1 again outperformed both competitors.
However, Anthropic’s model retained its lead on benchmarks such as Toolathlon-Verified, OSWorld-Verified, SWE-Bench Pro and CharXiv Reasoning, while GPT-5.5 topped Terminal-Bench 2.1, DeepSWE 1.1 and BabyVision.
Meta also highlighted improvements in computer-use capabilities. The company said Muse Spark 1.1 can automate workflows across applications, deciding whether to generate scripts or interact with graphical interfaces depending on the task.
The company also positioned the model for software development. Meta said Muse Spark 1.1 can diagnose bugs, implement features in large codebases and perform code migrations. It added that the model supports agentic coding workflows, including planning mode, goal conditioning, sub-agent delegation and context compaction.
Beyond coding, Meta said Muse Spark 1.1 can combine visual understanding with reasoning and tool use. The company said the model can generate visual-to-code outputs, create image and video descriptions, and execute multimodal workflows.
On safety, Meta said it evaluated the model under its Advanced AI Scaling Framework across chemical and biological risks, cybersecurity and loss-of-control scenarios.
The company said Muse Spark 1.1 remained within its deployment thresholds and showed resistance to jailbreak attempts, prompt injection attacks and developer prompt attacks.
ALSO READ: Alteryx Inspire 2026: Three Questions Every Data Leader Should Take to Orlando
