Gemini 4 Carbon hits new SOTA on Riemann-Bench with 0.8 Pass@1, beating GPT-5.6 Sol
NielsRogge · x · 2026-10-31
Papers with Code reports a new state of the art on the Riemann-Bench math benchmark: Google DeepMind's Gemini 4 Carbon scores 0.8 Pass@1, ahead of GPT-5.6 Sol at 0.74 and GPT-6 Astra at 0.72. Claude Opus 5.5 follows at 0.7, Claude Opus 5 at 0.68, and GPT-5.6 Terra at 0.66.
More from Models
- OpenAI's $100 Pro looks like a steal vs Anthropic's $200 Ultra: doubled quotas, no 5-hour window — gethackteam · 2026-10-31
- ChatGPT desktop voice mode still slow and unreliable, hands-on test finds — koltregaskes · 2026-10-31
- Google fixes Antigravity IDE to add Argon; early user calls it a good model — zacharynado · 2026-10-31
- Mistral open-sources Le Chonk, tops FrontierCode among open models — _AndrewZhao · 2026-10-31
- Blogger predicts OpenAI major release as Anthropic preps Fable 5.5 and new Opus — kimmonismus · 2026-10-31
- DeepMind: RL on simulated patients cuts AI psychiatry missed risk escalation from 30.5% to 12.9% — rohanpaul_ai · 2026-10-31