Publication date: July 10, 2026
Meta released Muse Spark 1.1 on July 9, 2026, the latest model from Meta Superintelligence Labs and the first Meta model available through a paid tier on the Meta Model API. Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks, with gains in tool use, computer use, coding, and multimodal understanding, and it can actively manage a 1 million token context window. API access is priced at $1.25 per million input tokens and $4.25 per million output tokens, marking the first time Meta charges developers for access to one of its models.
What is Meta Muse Spark 1.1?
Muse Spark 1.1 is a multimodal reasoning model from Meta Superintelligence Labs, the research group led by Alexandr Wang. It is an upgrade to the original Muse Spark and is designed for agentic work: planning and orchestrating actions across external apps, services, and tools rather than answering single prompts in isolation. Meta positions the model for personal agentic tasks that require multi-step planning, tool use, and computer use.
The model is multimodal and handles coding, video captioning, and reasoning across text and images. Meta reports that Muse Spark 1.1 can actively manage its 1 million token context window, zero-shot generalizes to new native tools and custom skills, and completes complex projects faster than the original Muse Spark. The release also marks a strategic shift for Meta, which moves from free model access to a paid developer tier and a new revenue stream through the Meta Model API.
Muse Spark 1.1 benchmarks and technical specs
Muse Spark 1.1 has a 1 million token context window and is built for agentic tool use. On MCP Atlas, a test of scaled tool use, it scored 88.1, ahead of Claude Opus 4.8 and GPT-5.5. On JobBench, which measures professional tool use, Muse Spark 1.1 scored 54.7 against Claude Opus 4.8 at 48.4 and GPT-5.5 at 38.3. It also led on Humanity’s Last Exam with tools at 62.1, ahead of Claude Opus 4.8 at 57.9, and on Finance Agent v2 at 57.2 against Claude Opus 4.8 at 53.9 and GPT-5.5 at 51.8.
On standalone coding, the picture is different. On Terminal-Bench 2.0, Muse Spark 1.1 scored 59.0, below GPT-5.5 at 82.7 and Claude Opus 4.8 at 65.4. On Meta’s internal coding benchmark, Muse Spark 1.1 scored 68.3, just behind Claude Opus 4.8 at 69.0 and just ahead of GPT-5.5 at 67.1. The results show a model tuned for multi-agent orchestration and tool use rather than solo, standalone coding tasks.
How does Muse Spark 1.1 compare to Claude Opus 4.8 and GPT-5.5?
Muse Spark 1.1 leads Claude Opus 4.8 and GPT-5.5 on agentic and tool-use benchmarks, including MCP Atlas (88.1), JobBench (54.7 vs 48.4 and 38.3), Humanity’s Last Exam with tools (62.1 vs 57.9), and Finance Agent v2 (57.2 vs 53.9 and 51.8). These evaluations reward planning, orchestration, and correct use of external tools, which is the area Meta optimized for.
On traditional standalone coding it trails both models. On Terminal-Bench 2.0 it scored 59.0 against GPT-5.5 at 82.7 and Claude Opus 4.8 at 65.4, and specialized long-horizon evaluations such as CyboBench show a wide gap to frontier coding models. For teams choosing a model, the practical takeaway is that Muse Spark 1.1 is strongest as an orchestrating agent that calls tools and coordinates across apps, and weaker as a solo code-generation model.
Muse Spark 1.1 availability and pricing
Muse Spark 1.1 is available in public preview on the Meta Model API, initially to developers in the United States. Pricing is $1.25 per million input tokens and $4.25 per million output tokens. Developers who sign up for the API receive $20 in free credits to test the model before moving to pay-as-you-go pricing. This is the first time Meta has charged businesses for access to one of its models, a shift from its earlier approach of releasing open or free-to-access models.
Conclusion
With Muse Spark 1.1, Meta Superintelligence Labs ships a multimodal agentic model with a 1 million token context window, leading tool-use benchmarks, and Meta’s first paid API tier at $1.25 input and $4.25 output per million tokens. Full details are in Meta’s announcement.
Official announcement: https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/
