Google DeepMind launches Gemini 4 Carbon, tops Artificial Analysis Intelligence Index two weeks after Argon, restricted to cyber defenders
2026-10-28
Google DeepMind announced Gemini 4 Carbon on Oct 28, positioned as state of the art on cybersecurity and long-horizon tasks, with lab-reported results of 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench. Artificial Analysis ranked Carbon #1 on its newly reweighted Intelligence Index v5 at a score of 59, ahead of Claude Opus 5.5, GPT-6 Astra, and all open-weight models, and #1 on LMArena Text. Initial access is limited to trusted cyber defenders through Google's Fairwind Program, with broader developer access promised "as soon as possible"; Artificial Analysis lists pricing at $4/$20 per million input/output tokens with a 50% launch discount.
Significance 5: A new model taking the top of two or more trusted independent leaderboards (Artificial Analysis Intelligence Index and LMArena Text) two weeks after its predecessor meets the rubric's bar for changing the landscape, even though restricted initial access tempers how quickly it affects day-to-day use.
Implications · machine-drafted, not owner judgment
If Carbon's independent scores hold once access widens, Google holds the #1 spot on the most-watched intelligence index for the first time this cycle — a direct test of the google-strategy-tax thesis, since Carbon (like Argon before it) launched cyber/enterprise-first rather than as a consumer feature. For the owner, a credible Google return to the frontier narrows the reasoning gap to gpt-5-2 (current hard-problems pick) and is worth revisiting at general availability; Google's restricted-access pattern on its last two flagship launches is itself a governance choice worth tracking.
- Whether Carbon reaches general developer availability on a similar timeline to Argon
- Independent replication of the SEC-bench Pro / ExploitGym cyber claims outside Google's own reporting
- Whether Carbon's coding-agent score (59 via Antigravity CLI) holds up in other harnesses, where it trails Claude Opus 5.5's 62
Sources
- primary Gemini 4 Carbon: Our next frontier model retrieved 2026-10-30
- curator [AINews] Gemini 4 Carbon tops leaderboards, launches for cyber defenders retrieved 2026-10-30
- curator What Google and Reddit's bad days have in common retrieved 2026-10-30
Affected axes
Hard reasoning · Coding · Agentic / tool use · Long context