Gemini 4 Carbon benchmark table: AA Index 59, leads most but not all agent evals
ammaar · x · 2026-10-29
Googler Ammaar confirms Gemini 4 Carbon ranks #1 on Artificial Analysis v5 (AA Index 59) and shares the full benchmark table:
- Agentic coding: SWE-sweep 9.6%, Terminal-Bench 4.0 65.2%, but FrontierSWE 2 (61.2%) trails GPT-6 Astra's 65.5%
- Computer use: Agents' Last Exam 77.4%; OSWorld-2.1 offline 81.5% below Claude Opus 5.5's 87.2%
- Long context: GraphWalks 96.2% (256k-1M), a wide lead
- Cybersecurity: best-in-class across SEC-bench Pro 88.2%, CWE-bench 71.0%, ExploitGym 54.8%
- Also tops Vending-Bench 2 ($16,484) and RiemannBench (80.0%)
More from Models
- repligate: Opus 3's uniqueness is consistently heroic, highest-good prioritization — repligate · 2026-10-29
- Gemini 4 Carbon benchmarks prompt TheZvi to declare Gemini 4 Argon 'was' near-frontier — TheZvi · 2026-10-29
- Preprint: Human peer reviewers caught only 27% of major errors, all 15 AI models beat them (53-96%) — anshulkundaje · 2026-10-29
- Giotto launches as 'European sovereign AI,' gets exposed as Kimi-based with benchmark EULA gag — TheZachMueller · 2026-10-29
- Google ships Gemini 4 Carbon two weeks after Argon, topping AA index at 59 — The Decoder · 2026-10-29
- Google Announces Gemini 4 Carbon, Its Next Frontier Model, Starting With Cyber Defenders — Snoo26837 · 2026-10-29