Meta Superintelligence Labs has released Muse Spark 1.3, an agentic coding model designed for long-horizon workflows rather than single-turn generation. The update is now available in Muse Code and the Meta Model API, featuring a 1M-token context window while keeping weights closed.

  • Internal comparisons show approximately 20% fewer tool calls and 25% fewer tokens compared to Muse Spark 1.2.
  • The model improves collaboration by asking clarifying questions and adapting to user preferences during long runs.
  • It achieves a score of 75.4 on DeepSWE v1.1, outperforming Claude Opus 5 and GPT-5.6 Sol.
  • Pricing remains unchanged at $1.25/M input and $4.25/M output tokens.

The release aims to reduce costs for agentic workloads by minimizing unnecessary turns and verbosity while maintaining strong performance on coding benchmarks.