⚡
Brief IA
›

Google Tests Carbon for Gemini 4, Argon on the Way

🤖 Models & LLM·Tom Levy·

Google Tests Carbon for Gemini 4, Argon on the Way

Google Tests Carbon for Gemini 4, Argon on the Way
⚡
Key Takeaways
1Google employees are testing Carbon, an internal version of Gemini 4
2Argon, the upcoming public version, must first go through the Fairwind program
3Google announces top performance and a focus on cybersecurity for Argon
💡Why it matters — Google is looking to catch up with its competitors in AI coding agents as the competition intensifies with Anthropic and OpenAI.
⚡Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

At Google, the Gemini 4 family is evolving behind the scenes. Employees are evaluating an internal version called Carbon, sometimes likened to Opus 5.5, while Argon must first go through the Fairwind program before a broader launch. Google highlights benchmark scores and a defensive focus on cybersecurity, while remaining silent about the exact status of Carbon.

Code Name, Uncertain Release, and Mandatory Passage Through Fairwind

The Barium-B model has been selected to be publicly presented under the name Argon, illustrating the usual difference between internal names and public names at Google. The status of Carbon remains undetermined: it could be an update under the name Argon or a distinct model within Gemini 4. The latest tested updates could reduce some performance gaps, but Google has not guaranteed their release. The company declined to comment. Officially, Argon must first be offered to partners in the Fairwind program, an initiative aimed at testing the cybersecurity vulnerabilities of new models, before any wider public rollout.

Official Promises of Argon: Benchmarks and Cybersecurity

Last month, Google indicated that Argon was achieving top results on various benchmarks related to coding and knowledge management, particularly in the legal and financial sectors. The company also emphasizes a priority given to defensive cybersecurity. The announced plan includes an initial availability through partners in the Fairwind program, which aims to test the vulnerabilities of new models before any general release.

Internal Testing: Carbon Under Trial and Compared to Opus 5.5

Google employees are testing a newer version of Gemini 4, internally referred to as Carbon, as evidenced by documents and screenshots. One employee noted that Carbon resembles Opus 5.5 for coding, while specifying that additional testing is needed. The early versions of Argon had been generally well-received by staff, although one employee reported delays on certain coding tasks, comparing it to Claude Opus 5. Discussions on internal channels confirm the testing activity: one message judges the update comparable to Opus 5.5, another considers Carbon to be very good, and yet another finds the new Gemini model to be quite pleasant, referencing the internal label Next assigned to new updates. Several branches of Gemini 4 have been tested: Argon, Barium, and Carbon.

Declared Timeline and Adjacent Announcements in Gemini

Google is preparing to make Argon publicly accessible, continuing the developments around Gemini 4 announced at the end of September. Public names generally do not correspond to internal code names. Additionally, the company introduced this week a workplace AI agent called Gemini Universal.

⚡

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.