Post-Cutoff
Search For AI agents

Google DeepMindGemini 4 familyPreviewReasoning modelReleased 28 Oct 2026

Gemini 4 Carbon

Google’s second Gemini 4 model. Not callable by the public at launch; this file will get the API id, endpoints, limits and cutoff when Google publishes them.

Price$2 in, $10 out per 1M tokens

Checked29 October 2026

No published training cutoff

At least55events since its release on 28 October 2026

0 historic, 11 major. 2 days of news.

Google DeepMind doesn’t publish a training cutoff, so we count from its release date. The real gap is larger.

Five things it has never heard of

How to call it

2 ways to use Gemini 4 Carbon.

Providers, model ids, endpoints and docs for Gemini 4 Carbon
ProviderModel idEndpointDocs
Google DeepMind Fairwind Program (a subset of cyber defenders from the week of 2026-10-28, then more trusted partners; public model id not published) deepmind.google
Gemini API / AI Studio, Gemini app, Vertex / Gemini Enterprise Agent Platform: not yet available; Google says 'soon' with no date blog.google

Pricing

Prices for Gemini 4 Carbon
Input$2 per 1M tokens
Output$10 per 1M tokens

As published: “per 1M tokens, introductory price announced by Google (cached input 95% off); after the introductory period $4.00 input / $20.00 output. No end date given. Not yet purchasable by the public (Fairwind partners only at launch)”.

Source: blog.google, checked 29 October 2026.

Specs

Input
Text, images, video and audio
Output
Text
Context window
1,000,000 tokens
Max output
Not published
Open weights
No
Licence
Proprietary
Training cutoff
Not published
Released
28 October 2026
Status
Preview
Type
Reasoning model

What stands out

What its maker marketed at launch, and what people found later, each with its source.

  • Top overall independent score at launch

    Artificial Analysis Intelligence Index v5: 59 (high), #1 of 52 on 2026-10-28, ahead of Claude Opus 5.5 at max (58) and GPT-6 Astra at max (57); #1 on HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. Second on AA’s Coding Agent Index (59 vs Opus 5.5 in Claude Code 62).

    Source: x.com

  • Cyber defense lifecycle (find, exploit, patch)

    Vendor-reported: SEC-bench Pro 88.2% (finding), ExploitGym 54.8% (exploitation; GPT-6 Astra 42.4%, Opus 5.5 34.5%), CWE-bench 71.0% (patching, Collinear leaderboard); more than 2x uplift in secure-code generation over Argon on internal tests. Self-computed with mitigations off.

    Source: storage.googleapis.com

  • Long-context reasoning

    Vendor-reported GraphWalks BFS 256K-1M F1 96.2% (Argon 84.2%, GPT-6 Astra 71.8%, Opus 5.5 66.8%).

    Source: x.com

  • Low hallucination rate with higher accuracy

    Artificial Analysis AA-Omniscience v1.1: score 54 (Opus 5.5 max 48), accuracy 58%, hallucination rate 11% (Argon 4%; Opus 5.5 max and GPT-6 Astra max 50%).

    Source: x.com

  • Top of Arena’s crowd-sourced Text Arena at launch

    Arena (ex-LMArena), Oct 28: #1 in Text Arena (1528, Argon 1520), #2 in Agent Arena ($1.80 median cost per task, 72% below #1 Claude Opus 5.5 Max), #4 in Code Arena WebDev (1768).

    Source: x.com

Notes

Successor to Gemini 4 Argon within the Gemini 4 family, announced 2026-10-28 as a gated release (cyber defenders first through the Fairwind Program). No public API id, docs page, model card or knowledge cutoff yet (ai.google.dev/gemini-api/docs/models/gemini-4-carbon and the deepmind.google model-card URL both returned 404 on 2026-10-28 and again on 2026-10-29); the id will probably be gemini-4-carbon, but that is unconfirmed. Context window 1M tokens and inputs text, image, video and speech per Artificial Analysis. AA measures ~100 output tokens/s and $7.76 per Intelligence Index task at the $4/$20 standard price; Vals AI: Vals Index 70.36% (#2, behind Claude Sonnet 5.5), Terminal-Bench 4.0 63.13%. Until it is generally available, agents should use gemini-4-argon.

Benchmark table from Google (Logan Kilpatrick’s launch post; methodology in the evaluation PDF). Rival scores are mostly the labs’ self-reported numbers.

Benchmark Carbon Argon GPT-6 Astra Opus 5.5 Fable 5.1
AA Index v5.0 59 50 57 58 49
Vals Index v2.2 70.4 69.2 65.0 70.2 —
SWE-sweep 9.6 4.3 9.5 3.6 2.9
FrontierSWE 2 61.2 55.0 65.5 62.3 56.3
Terminal-Bench 4.0 (AA) 65.2 57.1 59.1 59.6 52.0
Agents’ Last Exam v1.1 77.4 69.0 73.1 71.6 —
OSWorld-2.1 (offline) 81.5 74.7 80.9 87.2 76.8
AutomationBench 53.9 51.3 41.4 42.5 31.4
Vending-Bench 2 ($) 16,484 13,718 15,515 9,235 5,422
Chartography 72.5 71.6 71.0 66.3 46.2
Blueprint-Bench 2 58.6 54.4 49.7 51.2 41.9
GraphWalks 256K-1M 96.2 84.2 71.8 66.8 65.0
RiemannBench 80.0 76.0 72.0 69.6 65.6
Terminal-Bench-Science 0.1 (AA) 64.8 52.9 63.3 59.0 43.3
CWE-bench 71.0 68.0 68.0 67.0 58.0
ExploitGym 54.8 42.0 42.4 34.5 —
SEC-bench Pro 88.2 83.4 85.4 — —