Meta’s Muse Spark 1.3 Tops DeepSWE Benchmark with 75.4%

AI at Meta ·

Key Info

Meta’s Muse Spark 1.3 has reportedly achieved first place on the DeepSWE benchmark with 75.4%, ahead of GPT-5.6 Sol and Fable 5. The model is rolling out today in Muse Code and the Meta Model API, with the max reasoning version expected after safety testing.

Highlights

  • Claims a state-of-the-art 75.4% score on DeepSWE v1.1, topping other notable models.
  • Available now through Muse Code and the Meta Model API, with broader access planned after safety checks.
  • Marks another rapid iteration in the Muse Spark series, following a busy stretch of releases.
Loading...