Woofun AI reports that the Million team has released ReactBench v1, a benchmark designed to assess whether AI programming agents can execute practical React development tasks without introducing defects. The suite comprises 51 real-world tasks sourced from open-source projects, evaluating functionality alongside over 400 rules covering errors, performance, accessibility, and code quality.

GPT-5.6 Sol achieved the highest comprehensive score of 43.1%, narrowly ahead of Fable 5 at 41.2%. Officials noted the marginal difference and stated that Sol’s superiority is not yet statistically confirmed.

However, when configured for maximum capability, Fable 5 incurs costs approximately 6.3 times higher than Sol. Across 4,455 new feature tests, models collectively introduced 1,194 React issues, with 77.5% classified as program errors or security vulnerabilities.