Gemini 4 Carbon Explained

Gemini 4 Carbon is Google's new frontier model, and only a subset of Fairwind cyber defenders can use it today. At Layer3Labs, we build launch guides for new artificial intelligence (AI) models and route client workloads across vendors, so we track access rules and pricing shifts as soon as labs announce them. Google announced Gemini 4 Carbon on October 28, 2026, and gives no date for developers or consumers.

Gemini 4 Carbon ranks #1 of 52 models on the Artificial Analysis Intelligence Index v5.0, with a score of 59. Google calls Gemini 4 Carbon its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. Google does not say Gemini 4 Carbon replaces Gemini 4 Argon, which launched about two weeks earlier and stays the generally available Gemini 4 model.

Who Can Use Gemini 4 Carbon Today

Only a subset of cybersecurity defenders accepted into Google's Fairwind Program can access Gemini 4 Carbon as of October 28, 2026. Google states that it will phase access gradually to additional trusted partners through Fairwind over time, but it provides no target dates for developers, enterprise accounts, or consumers.

The public Gemini API documentation does not contain a Gemini 4 Carbon model identifier. A review of the Gemini API models page, updated October 22, 2026, shows that gemini-4-argon remains the sole listed Gemini 4 model ID.

Third-party listings present conflicting information regarding availability. For instance, Artificial Analysis lists Gemini 4 Carbon (High) as available via API through 1 provider at a post-introductory rate of $4 input and $20 output per 1M tokens. This entry contradicts Google's direct statement that access remains restricted to Fairwind participants.

The Fairwind Program prioritizes specific sectors: national cyber authorities, government bodies, critical infrastructure operators across healthcare, telecommunications, energy, and financial networks, as well as core technology platforms. Academic research laboratories focused on defensive evaluation may also apply through Google's interest form. Google vets every applicant organization and conducts background checks before granting model access. The live Fairwind documentation itself still describes access in terms of Gemini 4 Argon.

What Google Says Gemini 4 Carbon Is

Google describes Gemini 4 Carbon as its frontier model built directly on Gemini 4 Argon, released roughly two weeks after Argon's debut. The company positions Carbon as its strongest model to date, with state-of-the-art cybersecurity and long-horizon capabilities.

Google confirmed that Carbon participates in the United States government's voluntary process for pre-release model testing.

Google does not say Carbon replaces Gemini 4 Argon. Argon remains Google's generally available Gemini 4 model. Google has not announced any other variants under the Gemini 4 family name.

Gemini 4 Carbon Benchmark Highlights

Artificial Analysis lists Gemini 4 Carbon first of 52 models on its Intelligence Index v5.0, with a score of 59 as of October 28, 2026. On the developer leaderboard WebDev Arena, Google reports that Carbon gained 89 Elo points to reach fourth position overall with a 1768 rating.

In our internal analysis of Google's published table, we counted the top score in each row and found Carbon highest of the five models on 15 of 17 benchmarks. Google compared Carbon against Gemini 4 Argon, GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1. You can inspect every published Gemini 4 Carbon score across all 17 evaluations in our benchmark guide, where you can recount the rows yourself.

Carbon does not hold every benchmark record. OpenAI's GPT-6 Astra leads on FrontierSWE 2 with a score of 65.5% compared to Carbon's 61.2%. On OSWorld-2.1, Anthropic's Claude Opus 5.5 scores 87.2%, surpassing Carbon's 81.5%. Across the 16 benchmarks where both Carbon and Claude Opus 5.5 reported results, Carbon recorded higher scores on 14 while Claude Opus 5.5 won 2.

Against Gemini 4 Argon, Carbon demonstrated improvements across all 17 evaluations. The largest gains came on ExploitGym, up 12.8 percentage points to 54.8%, and on GraphWalks across 256k to 1M tokens, up 12.0 points to 96.2%. SWE-sweep more than doubled, from 4.3% to 9.6%.

Gemini 4 Carbon Price at Launch

Google announced that Gemini 4 Carbon will carry an introductory launch price of $2.00 per 1 million input tokens and $10.00 per 1 million output tokens. This matches the temporary launch rate established for Gemini 4 Argon. Cached input tokens receive a 95% discount during this introductory period, reducing the cached input cost to $0.10 per 1 million tokens.

Following the introductory period, standard pricing increases to $4.00 per 1 million input tokens and $20.00 per 1 million output tokens. Google has not published an expiration date for the introductory pricing window, nor has it confirmed whether the 95% prompt cache discount continues after rates adjust to $4.00 and $20.00. If the discount continues, cached tokens would cost $0.20 per 1 million tokens, but Google has not formally documented that figure.

Google has not published rate limits, output token maximums, or long-context pricing tiers for Carbon. For detailed token cost models across small and large prompt sizes, read our Gemini 4 Carbon pricing guide.

By comparison, Claude Opus 5.5 pricing stands at $4.00 input, $20.00 output, and $0.20 cached per 1M tokens in general availability. GPT-6 Astra charges $10.00 input, $50.00 output, and $1.00 cached per 1M tokens for prompts up to 272k tokens, rising to $20.00 input and $75.00 output above that threshold.

Safety Results Google Published

Google's announcement includes one prompt-injection result, from Gray Swan. On the Gray Swan evaluation of indirect prompt-injection attack success rates, Gemini 4 Carbon scored 0.0%, compared to a 0.7% attack success rate recorded for Gemini 4 Argon.

Google also reported internal safety findings that have not undergone external verification. Google stated that Carbon delivered more than a 2x uplift in its success rate for generating secure code over Argon. In addition, Google reported that Carbon cuts unsupported claims by approximately 2x compared to Argon.

Google says Gemini 4 Carbon's external cybersecurity benchmarks were evaluated without guardrails, so those scores do not show how Carbon performs with its guardrails on.

What to Use until Gemini 4 Carbon Opens

Teams requiring immediate frontier model capacity should build on Gemini 4 Argon rather than waiting for Gemini 4 Carbon access. Gemini 4 Argon is generally available on the Gemini API as gemini-4-argon, at the same introductory $2.00 input and $10.00 output per 1M tokens through November 30, 2026. You can review the trade-offs in our comparison of Gemini 4 Carbon vs Gemini 4 Argon and follow setup instructions in our guide on how to use Gemini 4 Argon.

Software teams shipping products this quarter should not wait for Gemini 4 Carbon. Google has given no date or public model ID for Gemini 4 Carbon, so build on a generally available model such as Gemini 4 Argon, Claude Opus 5.5 or GPT-6 Astra.

Our deployment recommendation would change immediately if Google publishes an official Gemini 4 Carbon model identifier on Google AI Studio, Vertex AI, or the standard Gemini API documentation. If your organization fits the Fairwind defensive cybersecurity criteria, apply through the Fairwind Program interest form. Otherwise, build on Gemini 4 Argon today and re-check the Gemini API models page for a Gemini 4 Carbon model ID.