Muse Spark 1.3 changelog

Arena October 22 snapshot: Muse Spark 1.3 (Max) 10th on text and 15th on WebDev

Arena's October 22, 2026 text leaderboard ranks Muse Spark 1.3 (Max) 10th at 1491 with 20,506 votes, ahead of Muse Spark 1.2 (xHigh) at 17th. The WebDev leaderboard of the same date ranks Muse Spark 1.3 (Max) 15th at 1657 and Muse Spark 1.3 (xHigh) 25th at 1620.

Vals AI Terminal-Bench 4.0: Muse Spark 1.3 Max at 24.8%, 24th

Running every model in the same Mini-SWE-agent harness, Vals AI lists Muse Spark 1.3 Max 24th on Terminal-Bench 4.0 at 24.75% for $6.65 per test, and its separately listed Muse Spark 1.3 entry 40th at 10.61%. On Vals AI's ProofBench, Muse Spark 1.3 Max scores 58.00% and Muse Spark 1.3 55.00%.

Epoch: Muse Spark 1.3 (max) scores 74% on FrontierMath tiers 1-3 and 46% on tier 4

Epoch benchmark data records Muse Spark 1.3 at max effort at 74.04% on FrontierMath tiers 1-3 and 46.34% on tier 4, from runs started September 18, 2026. Runs at xhigh effort, started September 16, scored 74.39% and 41.46%. Epoch's Capabilities Index lists Muse Spark 1.3 at 157.21 and Muse Spark 1.2 at 155.05.

Muse Spark 1.3 max reasoning becomes available in Muse Code and the Meta Model API

Two days after launch, Meta announced that Muse Spark 1.3 with max reasoning is available in Muse Code and the Meta Model API. Alexandr Wang wrote that Meta had "just publicly released" it and recommended trying it even after high or xhigh, citing "significantly stronger coding and agentic performance".

Who is affected:

Developers using Muse Spark 1.3 on the Meta Model API or in Muse Code.

What to do:

Try the max reasoning setting for hard coding and agent tasks.

Muse Spark 1.3 released on Muse Code and the Meta Model API

Meta released Muse Spark 1.3 with better agentic and coding performance, more reliable long instruction following and clarifying questions when prompts are ambiguous. The max reasoning setting followed on September 4.