Muse Spark 1.1

Meta Platforms Inc. today launched Muse Spark 1.1, a new flagship large language model optimized to power multi-agent automation workflows, the company announced. The model is available in the Meta AI chatbot service and via the Meta Model API, which is in public preview and enables developers to embed the large language model in custom software. Multi-agent workflows typically involve a main agent that plans tasks and subagents that execute instructions. Meta said Muse Spark 1.1 can detect and respond to mid-task developments that require changes to the project plan. The model addresses context limit challenges with a context compaction mechanism that compresses data generated by agents while preserving important details, allowing it to retrieve information from earlier work. The model has a context window of 1 million tokens. In an internal test, Meta engineers had the model generate a chat application, take screenshots of the interface, identify technical issues and fix the code. Muse Spark 1.1 scored 72.2 on the Vibe Code Bench v1.1 programming benchmark, more than 50 points ahead of Meta's previous flagship model, and achieved a nearly 18% higher score on the SWE-Atlas Codebase QnA test. The model can also generate e-commerce listings from product videos and place restaurant orders. Reuters reported today that Meta plans to boost data center capacity to 14 gigawatts next year around its Iris AI chip, which is likely the MTIA400 processor previewed in March.