Gemini 4 Neon benchmarks leak: leads Claude Opus 5.5 and GPT-6.2 Sol on most evals
casper_hansen_ · x · 2026-11-20
The author argues Gemini 4 marks Google's real return to the AI frontier — a full release rather than a preview — with a genuine step up in agentic coding, crediting Sergey being back in founder mode.
The attached comparison table shows Gemini 4 Neon beating Claude Opus 5.5 and GPT-6.2 Sol on most benchmarks: 95.2% on GraphWalks long-context (vs 66.8%/80.1%), 90.5% on SEC-bench Pro, 75.8% on Agents' Last Exam, 71.3% on HLE-Diamond, and 39.0% on SWE-sweep with provider harness. It still trails Claude Opus 5.5 slightly on Terminal-bench 4.1 (65.2% vs 65.8%) and FrontierCode 2.0, and loses Chartography to GPT-6.2 Sol — strong overall lead, but Anthropic retains an edge on select coding evals.
Related event: Leaked Gemini 4 Neon Benchmark Chart Shows Leads Over Opus 5.5 and GPT-6.2(3 posts)→
More from Models
- Is Google back? Neon and Carbon spark talk of stockpiled frontier models — almmaasoglu · 2026-11-20
- User finds ChatGPT gets skittish about sodium bicarbonate when it knows you take Adderall — repligate · 2026-11-20
- Gemini 4 Neon flunks Runescape Bench: ambitious strategies, terrible time management — morgymcg · 2026-11-20
- Open-source cancer model from JHU and Stanford hits every tumor across 34 public CTs — MaziyarPanahi · 2026-11-20
- Fan makes an AI video showing how much ~30B Qwen models improved in 28 months — vanstriendaniel · 2026-11-20
- moyix on LLMs' worst writing habit: "disclaiming things no one ever claimed" — moyix · 2026-11-20