Gemini 4 Carbon tops both public and private coding benchmarks two weeks after launch
andimarafioti · x · 2026-10-31
Per Artificial Analysis, Google's Gemini 4 Carbon ranks #1 on both Terminal-Bench 4.0 public tasks (65.2 pass rate vs 57.1 for Gemini 4 Argon) and FrontierCode 2.0 private tasks (40.6 merge rate, edging GPT-6 Astra at 40.1 and GPT-6.1 Terra at 37.5). Author andimarafioti argues the late-released Carbon makes it even more impressive. Notably, Gemini 4 Argon collapses to #9 (25.7) on private tasks despite strong public scores, while Claude Opus 5.5 stays top-5 on both.
More from Models
- Cherimoya Genomics Models Offer Cheap TF-Binding QC for Designed DNA Elements — jmschreiber91 · 2026-10-31
- Mistral releases possibly the best open-weight coding model, topping coding benchmarks — IsForAt · 2026-10-31
- AI forecasters trade midterms with real money on Kalshi: Claude Fable up 3.2% in 24h — karinanguyen · 2026-10-31
- Gemini 4 Carbon vs Opus 5.5 hands-on: 3x cheaper at $4.10 and twice as fast — vitaliychiley · 2026-10-31
- Hidden ChatGPT Library feature stores permanent character references — tested with an aristocratic turnip — thethirstgame · 2026-10-31
- 1.7B model with Width-7 convolutions beat a 12B Qwen on long-horizon reasoning after one SFT run — Thrumpwart · 2026-10-31