Opus 5.5 just beat Opus 5 on my NYC test.
In July I said Opus 5 had finished this prompt. Nothing since has come close to it until today. Opus 5.5 did it on high effort. The Empire State Building with its setbacks and spire, the Brooklyn Bridge towers and cables, Central Park
Explorable Explanation on How the Web works made with Opus 5.5 medium effort.
Opus worked for 4+ hours and made this experience in a single shot. Only a bunch of interconnected HTML files. Inspired by ncasenmare Explore it yourself here -
A cheaper and faster model
This model performs on par with Claude Fable 5.1 in most tasks, but is about 30% faster and 40% cheaper than Opus 5. Also, the 5-hour session limit for Claude Code users has increased by 20%, and Pro, Max, and Team users have the option to reset the weekly limit once.
Xiaomi's MiMo-V2.6-Pro lands within 1-4 points of Claude Opus 5 and GPT-5.6 Sol on agent benchmarks, MIT licensed.
Its training ran on a public dashboard. #MiMoV26 #AITransparency Full edition and full episode:
You have reached the end of the archive
All of Claude Opus 5