# Gemini 4 Carbon, explained

> Gemini 4 Carbon is Google's newest frontier model, announced October 28, 2026, built for cybersecurity and long engineering jobs. Access is gated: first to a subset of cyber defenders through Google's Fairwind Program, then to more trusted partners. Developers, businesses, and Gemini app users have no published date yet. Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input at 95% off; standard pricing is $4 and $20, with no published end date for the introductory period. Full guide: https://lab.awaara.cn/gemini-4-carbon/

## Key facts

- Announced: October 28, 2026, by Google DeepMind (Koray Kavukcuoglu, SVP and Chief AI Architect).
- Position: second Gemini 4 model, two weeks after Gemini 4 Argon; Google calls it its strongest model.
- Artificial Analysis Intelligence Index v5.0: 59, first, ahead of Claude Opus 5.5 (58), GPT-6 Astra (57), Gemini 4 Argon (50).
- Security benchmarks (Google-reported, run without guardrails): SEC-bench Pro 88.2% (finding vulnerabilities), ExploitGym 54.8% (exploitation), CWE-bench 71.0% (patching).
- WebDev Arena: 1769 Elo, fourth place, up 89 points.
- Internal claims (not independently verified): secure code generation more than twice Argon's rate; Chrome team high-quality patches more than 10 times better than another lab's unnamed model; high-severity misalignment alerts down almost 80%; 0.0% cheating on Terminal-Bench 4.0; 0.0% attack success on Gray Swan indirect prompt-injection benchmark (Argon: 0.7%).
- Pricing: introductory $2/$10 per million tokens, cached input $0.10; standard $4/$20, cached input $0.20 (assuming the 95% discount continues). Introductory end date: not published.
- Access order: subset of cyber defenders, then more trusted partners via Fairwind, then developers, businesses, and consumers. No dates for later groups.
- Not published as of October 30, 2026: parameter count, context window, general availability date, introductory pricing end date, public API model ID.
- Open weights: no. Closed Google model only.
- Caution: independent coverage reports Carbon trails Opus 5.5 on agentic coding and that its hallucination rate rose from Argon's 4% to 11% on one factual-knowledge test.

## Optional

- Independent explainer, not affiliated with Google. Prices and availability change; check Google's announcement before buying.