Meta Launches Muse Spark 1.1 Multimodal Reasoning Model
Original: Muse Spark 1.1
Why This Matters
Meta's entry into the agentic AI model market with API access signals intensifying competition with OpenAI and Anthropic in the enterprise AI space.
Meta Superintelligence Labs released Muse Spark 1.1 on July 9, 2026, a multimodal reasoning model with a 1M-token context window. It is available in 'Thinking' mode on the Meta AI app and meta.ai, with API access via a new Meta Model API public preview.
Meta Superintelligence Labs unveiled Muse Spark 1.1, a significant upgrade to its predecessor Muse Spark, designed for agentic tasks including tool use, computer use, coding, and multimodal understanding. The model supports a 1 million token context window and can orchestrate multi-agent systems to reduce end-to-end latency. As a main agent, it can gather context, plan, and delegate tasks to parallel subagents; as a subagent, it knows when to escalate to the main agent. In computer-use scenarios, Muse Spark 1.1 can write automation scripts when faster and click interfaces directly when simpler, maintaining context across extended multi-application sessions. On coding benchmarks, it shows substantial improvements on real-world tasks involving large codebases, including bug diagnosis, feature implementation, and large-scale code migrations. The model performs well in agentic coding setups supporting planning mode, goal conditioning, and context compaction. Internally, Meta reports that Muse Spark 1.1 significantly outperforms Muse Spark on Meta Internal Coding Bench and is competitive with leading alternatives. The release coincides with the launch of Muse Image. Developers can access the model through a new public preview of the Meta Model API.