Meta's Muse Spark 1.2 Rises to #14 on Code Arena WebDev Benchmark
Meta AI's Muse Spark 1.2 (xHigh) has climbed to rank #14 on the Code Arena: WebDev benchmark with 1,545 points. This version shows a notable improvement over Muse Spark 1.1, which previously held the #18 spot. The advancement demonstrates Meta's progress in developing native multimodal reasoning models capable of handling complex web development and coding tasks. Higher rankings on user-voted benchmarks like Code Arena indicate better real-world performance in generating functional front-end code. Muse Spark 1.2 saw its most significant gains in the HTML category, jumping from #24 to #8, and Gaming, rising from #23 to #13. The model features a 1-million-token context window and supports multimodal inputs including text, images, video, audio, and PDFs.
## BACKGROUND
Muse Spark is a reasoning model family built from scratch by Meta's Superintelligence Labs, rather than being an iteration of the Llama series. The Code Arena: WebDev benchmark evaluates AI models on front-end web development tasks using human preference Elo ratings based on anonymous side-by-side comparisons.