Gemini 4 Carbon not listed in API models

On October 30, the list of Gemini API models includes only Gemini 4 and Argon, which was released on October 13. Gemini 4 Carbon, which Google presented two days earlier as its "most powerful" model, is not listed, and the pricing page does not allocate any line to it.

Gemini 4 Carbon leads in AI index

Despite its absence from the API list, Gemini 4 Carbon is the top performer in the Artificial Analysis Intelligence Index v5, with 59 points. Google does not sell it to any enterprise through self-service, and Artificial Analysis states it is "available via API through a provider," the Google API, which the Google documentation does not describe.

Access to Gemini 4 Carbon is restricted

The announcement post, signed on October 28 by Koray Kavukcuoglu, senior vice president of Google DeepMind and chief AI architect of Google, describes access through circles. Carbon is available from the week of its launch to "a subset of cybersecurity defenders," and will gradually be made available to "other trusted partners" through the Fairwind Program. Developers, businesses, and the general public will come next: Google says it will continue "to gather feedback and adjust safeguards" before opening the model "as soon as possible." No date is given.

Google mentions a voluntary testing process

The same post indicates that Google participates in the "voluntary government process" of testing models before publication, without naming the organization. The Center for AI Standards and Innovation (CAISI), affiliated with NIST, announced on May 5, 2026, agreements with Google DeepMind, Microsoft, and xAI to evaluate models "before they are accessible to the public." Google does not state whether this is the process it refers to, and NIST does not describe these agreements as voluntary.

Gemini 4 Carbon outperforms other models

Artificial Analysis published the fifth version of its index on October 27. On that day, Claude Opus 5.5 led with 58 points, ahead of GPT-6 Astra at 57. It lost the top spot the next day. On the ranking consulted on October 30, Carbon, in its "high" reasoning setting, surpassed it with 59 points, one point more than Opus 5.5 and two points more than GPT-6 Astra, both measured in their maximum settings. Its profile ranks it first among 57 comparable models.

Google highlights security performance

Google also claims records on selected security tests, including 88.2% on SEC-bench Pro, dedicated to vulnerability hunting, and specifies that these measurements were conducted "without safeguards." These are precisely the protections that the company is still adjusting before any broad release.

Other companies also restrict access

Anthropic and Mistral also keep models in closed circles. On October 27, the Frontier Red Team of Anthropic described in a post the critical flaws found by "an internal research model, not published," in browsers and operating systems. In Chrome bulletins, a researcher from Anthropic signed a flaw out of ten fixed in October, without it being possible to link these credits to this model. The post says nothing about a possible commercialization of this model.

Claude Mythos Preview has limited access

Claude Mythos Preview follows a more explicit logic. Anthropic has reserved it since April 7, 2026, for members of the Project Glasswing: Amazon Web Services, Apple, Google, Microsoft, and seven other launch partners, and then more than forty organizations that maintain critical software. The page states: "we do not plan to make Claude Mythos Preview generally available, but our goal is, in the long term, to allow our users to deploy Mythos-class models securely at scale." After the research phase, participants will pay $25 per million input tokens and $125 for output tokens.

Mistral AI restricts access to a cybersecurity variant

Mistral AI, which published the weights of Mistral Large 4 on October 29, has reserved a cybersecurity-specialized variant for partners such as Orange Cyberdefense. Its post submits this access preview to "rigorous verification and contractual guarantees," with tests limited to systems owned or explicitly authorized. None of these four models have a general release date. Carbon is the only one to occupy the top spot on a general index.

Pricing for Gemini 4 Carbon will double

A call rate that will double, like that of Argon. Google launches Carbon at the same introductory price as Argon: $2 per million input tokens, $10 for output, and a 95% discount on input tokens served from the cache, i.e., $0.10. A note at the bottom of the post warns that after the introductory period expires, the rate will increase to $4 and $20. The note does not date this expiration. Artificial Analysis already lists Carbon at $4 and $20, based on the Google API. Argon shows how the mechanism works. The note on Argon's post, however, dated this expiration: November 30, 2026.

Carbon's full price compared to other models

At the full price, Carbon will cost as much as Claude Opus 5.5, priced at $4 and $20 according to the price list from ActuIA's API, and two and a half times less than GPT-6 Astra, at $10 and $50. The only rate on which a company can build a Carbon budget is the $4 and $20 rate: Google does not date either the model's opening to customers or the end of its introductory period.

Fairwind Program offers limited access

On October 30, the Fairwind Program page, which claims "more than 650 partners," still offers Gemini 4 Argon in a version equipped with "cybersecurity capabilities," as well as the code correction agent CodeMender. Carbon is not listed there.