Google launches Gemini 4 Carbon: SOTA in cybersecurity, priced $2/$10 per million tokens
rohanpaul_ai · x · 2026-10-29
Google DeepMind officially unveiled Gemini 4 Carbon, a new frontier model arriving just two weeks after Gemini 4 Argon and beating GPT-6 Astra and Claude Opus 5.5 on Artificial Analysis.
Key points
- Cybersecurity SOTA: 88.2% on SEC-bench Pro (vuln finding), 71.0% on CWE-bench (patching), 54.8% on ExploitGym (exploiting). Access rolls out first to cyber defenders and Fairwind Program partners, then developers and businesses.
- Long-horizon engineering: Google says Carbon can work on engineering problems for days, root-causing subtle bugs across large codebases.
- Knowledge work: tops AutomationBench (Zapier) at 53.9% and Vending-Bench 2 with a $16,484 bank balance.
- Long context: 96.2% on GraphWalks (256K–1M, BFS F1) vs Argon's 84.2%.
- Pricing: introductory $2/$10 per 1M input/output tokens, later doubling to $4/$20.
Related event: Google Launches Gemini 4 Carbon, Topping Artificial Analysis Index(112 posts)→
More from Models
- Unsloth quants run 1T-param Mistral Large 4 on Mac Studio: 1-bit keeps 91% accuracy at 223GB — danielhanchen · 2026-10-29
- Users report Qwen3.8-Flash-Next produces verbose, dense, metaphor-laden docs — walden42 · 2026-10-29
- Mistral's 1T-parameter Le Chonk lands on Hugging Face with llama.cpp support — Disastrous-Work-1632 · 2026-10-29
- Mistral Large 4 'Le Chonk' lands on Hugging Face: 1T-param MoE, 49B active, 1M context — ariG23498 · 2026-10-29
- Mistral open-sources Le Chonk, tops FrontierCode among open models for agentic coding — MistralAI · 2026-10-29
- Meta and Oxford's OverSeer lifts multi-agent world-model consistency on Minecraft to 60.2% from 0–12.5% baselines — armand_ruiz · 2026-10-29