Recent internal testing of muse-spark-1.4-internal, evidenced by a November 13 GitHub pull request logging real calls with over 33,000 input tokens, signals that Meta Superintelligence Labs has advanced the next closed-weight version beyond the September 2 release of Muse Spark 1.3. This follows a pattern of roughly monthly point releases focused on agentic coding efficiency, reduced tool calls, and multimodal reasoning, though the October open-weights Spark Open shifted some resources. Traders view the leak as accelerating timelines toward a late-November or December public launch via the Meta Model API or Muse Code, tempered by Meta’s history of controlled rollouts and competing priorities like the higher-end Muse Wonder line. Key catalysts ahead include potential benchmark leaks or developer conferences that could confirm readiness or reveal delays.