Meta Superintelligence Labs has released Muse Spark 1.1, a multimodal reasoning model designed for agentic tasks that significantly improves performance in tool use, computer interaction, and coding compared to its predecessor.
- The model features a 1 million token context window and can orchestrate multi-agent systems to optimize end-to-end latency.
- It excels at computer-use workflows by adapting to evolving requirements and automating actions across multiple applications with minimal human intervention.
- Coding capabilities allow it to diagnose bugs, implement features in enterprise-grade systems, and handle complex code migrations.
- Multimodal strengths include visual-to-code artifact generation, ultra-descriptive captioning, and reasoning over video and audio inputs.
- Safety evaluations indicate strong resistance to jailbreaks and prompt injection, with reduced hallucination rates and lower sycophancy.
Muse Spark 1.1 is available via the new Meta Model API in public preview and within the Meta AI app, aiming to provide a complete agentic foundation for developers.