Introducing Muse Spark 1.1

On July 9, 2026, Meta officially released Muse Spark 1.1, an upgraded version following Muse Spark in April, which Meta calls a "major upgrade." The model is designed for agentic tasks, coding, and computer use, and is now available in public preview to developers via the Meta Model API for the first time.

Muse Spark 1.1 supports a 1 million token context window and can actively manage context, including memory, retrieval, and information compression. Meta states it is trained to orchestrate multi-agent systems, acting as a lead agent that can plan and delegate tasks to parallel sub-agents, and as a sub-agent that can follow tasks and escalate when appropriate.

The Meta Model API is priced at $1.25 per million input tokens, $4.25 per million output tokens, and $0.15 per million cached input tokens. This price is significantly lower than OpenAI's GPT-5.5 and Anthropic's Opus 4.8 (which range from $25 to $50 per million output tokens), and even lower than xAI's newly released Grok 4.5.

The Decoder analysis suggests that this aggressive pricing could pressure pure AI labs like OpenAI and Anthropic, as they rely on high profit margins to cover losses and support valuations. Tech giants like Meta and Google can treat the API as an ecosystem entry point without needing short-term profitability. Meanwhile, Chinese open-source models are also pushing prices down from another direction.

Meta claims Muse Spark 1.1 scores 88.1 on the MCP Atlas benchmark and 62.1 on Humanity's Last Exam, leading Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. However, on SWE-Bench Pro, Opus 4.8 leads with 69.2, while Muse Spark 1.1 scores 61.5.

The Decoder notes that Meta's benchmark results still need verification in production environments, and actual token efficiency may affect total cost. In the independent VALS-AI benchmark, Muse Spark 1.1 ranks fourth overall, but jumped 36 places on the "Vibe Code Bench" compared to its predecessor.

Meta states that it conducted extensive safety evaluations under the Advanced AI Expansion Framework before deployment, and that Muse Spark 1.1 operates within safe bounds across all frontier risk categories (chemical and biological, cybersecurity, and loss of control).

The API is currently limited to US developers, and each new account comes with $20 in free credits. The model is not open-sourced, continuing a closed-source strategy, contrasting with the previous open-source approach of Llama.