Brief IA
›

Google Launches Gemini 4 Carbon, First on SEC-bench Pro

AI Tools·Tom Levy·

Google Launches Gemini 4 Carbon, First on SEC-bench Pro

Google Launches Gemini 4 Carbon, First on SEC-bench Pro
Key Takeaways
1Google announces Gemini 4 Carbon, touted as more efficient in cybersecurity, coding, and agentic tasks
2The model ranks first on SEC-bench Pro, ExploitGym, and the Artificial Analysis Intelligence Index v5.0
3Access is initially reserved for defense companies, with an extension planned through the Fairwind Program
4The announced pricing is $2 per million input tokens and $10 per million output tokens
Why it matters : Gemini 4 Carbon represents a claimed advancement by Google in vulnerability detection and secure code generation, with top performance on several key industry benchmarks.

Google introduces Gemini 4 Carbon, a model touted as more efficient in coding, agentic tasks, and cybersecurity. The company claims to lead in several benchmarks, including SEC-bench Pro and ExploitGym, and describes a significant improvement in cyber defense. Access is currently limited to defense actors, with an extension planned for partners of the Fairwind Program. The announced pricing mirrors that of Argon.

Carbon aims for multi-day tasks and reduces unsupported claims

Google states that it designed Gemini 4 Carbon for long missions. According to its teams, the model can operate for several days to overcome bottlenecks in complex engineering tasks that involve multiple steps. Google specifies that Carbon better avoids shortcuts, is more resistant to prompt injection attacks, and collaborates more effectively with users. The company also claims that the model offers greater transparency and honesty, halving the number of unsupported claims compared to Argon.

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.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Benchmarks: Google claims the lead on AAI Index v5.0 and SEC-bench Pro

Google highlights Gemini 4 Carbon's top position on the Artificial Analysis Intelligence Index v5.0, ahead of Claude Opus 5.5 and GPT-6 Astra. The company also notes that Carbon ranks first on SEC-bench Pro, a benchmark for detecting real bugs, as well as on ExploitGym, which measures the ability to turn a vulnerability into an exploit. Gemini 4 Carbon has also outperformed GPT-6 Astra on SEC-bench Pro. Google describes a "significant leap" in cyber defense, with better vulnerability detection and safer code generation. More broadly, Carbon is presented as surpassing models from OpenAI and Anthropic on certain benchmarks, and as delivering notable gains in coding, agentic tasks, and cybersecurity.

Restricted access, extension via Fairwind, and announced pricing

Gemini 4 Carbon is currently accessible only to a selected group of cybersecurity defense companies. Google plans to gradually extend this access to other partners as part of its Fairwind Program. The model was announced on a Wednesday. Pricing is set at $2 per million input tokens and $10 per million output tokens, the same introductory price as Argon. Carbon is presented as an improvement over Argon, which was launched earlier this month.

Brief IA — L'actualité IA en français

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