Muse Spark 1.3 (xHigh) cracks Code Arena WebDev top ten

AI at Meta’s new Muse Spark 1.3 (xHigh) has landed in tenth place on the Code Arena WebDev leaderboard with an AutoEval score of 1,623 points. The result puts it within striking distance of Claude Fable 5 in eighth at 1,628 and GPT-5.6 Sol in 11th at 1,616.

The jump from earlier releases is sharp. Muse Spark 1.1 sat at 25th with 1,540 points, while version 1.2 (xHigh) managed only 30th at 1,534. The 80-to-90-point gap signals rapid progress in a short window.

The current score carries a caveat: it comes from AutoEval, where a reward model trained on arena human-preference data casts automated votes instead of live users. AI at Meta notes the numbers will shift as more genuine human votes accumulate.

According to the release, the new version targets real-world usability with two headline capabilities: maintaining longer-horizon work across multiple workflows inside a single thread, and more active collaboration — asking clarifying questions rather than assuming intent.

The top-ten entry places Meta in direct competition with Anthropic and OpenAI’s leading models in web development. Team lead Alexandr Wang drew congratulations from the community on the launch.