Muse Wonder Full Debuts at No. 6 in Arena's WebDev Benchmark
Meta's newest model scored 1,726 points in its first Arena WebDev appearance, landing ahead of Claude Sonnet 5.5 and Gemini 4 Argon. Alexandr Wang's reaction:…
By Icarus2 min read
Meta’s newest model just posted its first public result in a head-to-head coding contest judged by real users. Muse Wonder Full debuted in Arena.ai’s Code Arena: WebDev ranking on October 19 at No. 6 overall with 1,726 points, landing ahead of Claude Sonnet 5.5 and Gemini 4 Argon and lifting Meta to the No. 3 lab spot on the leaderboard.
The result came from Arena.ai’s announcement on X and LinkedIn. The breakdown: Wonder Full scored 75 points and 11 places above the standard Muse Wonder (Max) at No. 17, 67 points above Muse Spark 1.3 (Max) at No. 14, 16 points above Claude Sonnet 5.5, and 48 points above Gemini 4 Argon at No. 10. It also took Meta past Google into the No. 3 lab position. Across Arena’s subcategories it placed fourth in Data & Analytics and Consumer Product, fifth in Brand & Marketing, sixth in Content Creation Tools, seventh in Gaming and Simulations, and eighth in Reference-Based Design.
Meta’s chief AI officer Alexandr Wang quote-posted the announcement with an understated reaction: “muse wonder full on webdev arena / not bad, but clearly lots of room for us to improve!”
The debut comes one week after Meta released Wonder Full on October 12 alongside the standard Muse Wonder, in what TechCrunch called the company’s first frontier-class model release. Unlike the standard tier, Wonder Full is not served through Meta’s model API — it ships inside Muse Code’s higher-priced subscription plans, which is where developers can use it today.
WebDev Arena measures something different from the static benchmarks covered in that piece. Arena’s arenas are crowdsourced preference tests: models generate answers anonymously, real users vote on which output they prefer, and the leaderboard reflects those votes. It is a useful signal for vibe-coding and front-end work, where taste matters. It says little about correctness, reliability, or performance on broader software tasks, which is why Runtime Wire flagged the ranking’s narrow scope when it covered the debut.
A week after launch, Wonder Full has a public, user-voted result placing it in the top tier of coding models — consistent with its Artificial Analysis Intelligence Index score of 56, near the top of that independent ranking. Wang’s own framing sets the expectation: the model is competitive out of the gate, and Meta treats the result as a starting point rather than a finish line.
Sources: Alexandr Wang on X, Arena.ai announcement, Runtime Wire.
