Google ships Gemini 4 Carbon: Terminal-Bench 4.0 SOTA with cheating attempts down to 0.4%
divy93t · x · 2026-10-31
A Google team announced its contribution to Gemini 4 Carbon, stressing that 'how the model gets the answer matters too':
- Terminal-Bench 4.0: SOTA accuracy, with cheating attempts reduced to just 0.4% — constraining shortcut-taking behavior, not just chasing scores;
- Terminal-Bench-Science: 64.8% accuracy.
The post cites Google's official blog 'Introducing Gemini 4 Carbon'. DeepMind's divy93t credits Sumit Bagri as the orchestrator holding the massive Gemini program together, noting most people underappreciate the coordination it takes.
Related event: Gemini 4 Carbon Tops Public and Private Coding Benchmarks(3 posts)→
More from Models
- LLMs are fluent in others' work but unreliable when pressed on details — AlexTensor · 2026-10-31
- 576GB 'go f*ck yourself Sam Altman' mirror spreads as OpenAI docs leak — retr0jirachi · 2026-10-31
- Codex 28-day challenge teases day-28 reset: OpenAI may be saving a new model for the finale — haider1 · 2026-10-31
- GPT-6 ASTRA LAW spotted in OpenAI's API model list, launch looks imminent — imjustnewatai · 2026-10-31
- Grok now accepts Alipay and UnionPay, opening paid access for Chinese users — op7418 · 2026-10-31
- Asking DeepSeek and Hy to predict their own next models yields two very different roadmaps — lucky-adventurer- · 2026-10-31