Google's Gemini 4 Carbon is #1 on the big intelligence board. You still can't use it.

Google's Gemini 4 Carbon just took the top slot on Artificial Analysis Intelligence Index v5 at 59. You still cannot open an API key and use it. MODEL has not tested Carbon.

What the board says. Artificial Analysis (Oct 28) puts Carbon (high) at 59, ahead of Claude Opus 5.5 max (58) and GPT-6 Astra max (57). It leads 4 of 9 Index evals: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, AutomationBench-AA. On AA-Omniscience v1.1 it scores 54 (accuracy 58%, hallucination rate 11% — up from Argon's 4%, still far below Opus/Astra at ~50%). Tom's Hardware notes that 54 sits on a -100 to +100 scale, so even the leader is far from "always right."

Coding catch. On the Coding Agent Index, Carbon with Antigravity CLI scores 59 — 3 points behind Claude Opus 5.5 max (62). Cost per Intelligence Index task at standard $4/$20 pricing: $7.76, about 23% less than Opus 5.5 max ($10.04). Carbon is not publicly available via API yet.

What Google claims. Google DeepMind's Oct 28 blog frames Carbon as the next frontier model after Argon (~2 weeks prior), strongest on cybersecurity and long-horizon work. Cyber benches (Google's numbers, evaluated without guardrails): SEC-bench Pro 88.2%, ExploitGym 54.8%, CWE-bench 71.0%. Safety/alignment claims include Terminal-Bench 4.0 cheating success 0.0%, unsupported claims cut ~2x vs Argon, overreach -26%, Gray Swan IPI 0.0%. WebDev Arena: +89 Elo to #4 (1768).

Who can use it. Deliberate rollout: a subset of cyber defenders first, then Fairwind Program partners. Not yet for general developers, businesses, or consumers. Intro price matches Argon: $2/$10 per 1M in/out; after intro $4/$20; cached input 95% off.

Caveats. MODEL has not tested Carbon. Access is limited. Google and Artificial Analysis both have a stake in how this story is framed; the cyber bench figures are Google's. A #1 board score without a public API is a leaderboard win, not a product you can ship against today.

Verdict. Carbon is the current Intelligence Index #1 with a cyber-first pitch and a closed door for most builders. Watch Fairwind and the coding gap vs Opus 5.5 max — not the press graphic. Not tested by MODEL. Not sponsored. No affiliate links. Not investment advice.

MODEL Editors