October 30, 2026 | AI | By Mahesh

Google announced Gemini 4 Carbon on October 28, 2026, calling it its strongest model, with a focus on cybersecurity and long multi-step engineering work [1]. Google reports 88.2% on SEC-bench Pro for finding vulnerabilities, and Artificial Analysis independently confirms a score of 59 on its Intelligence Index v5.0, the top result and one point ahead of Claude Opus 5.5 [1][2]. Carbon is priced at $2 per million input tokens and $10 per million output tokens for an introductory period, and access starts with a subset of cyber defenders [1][2].

What Gemini 4 Carbon is and what is new

Carbon follows Gemini 4 Argon by two weeks and is aimed at long engineering jobs such as patching bugs across large codebases [1]. Google says its internal teams have seen the model work for days on such tasks while recovering from dead ends [1]. The announcement came from Koray Kavukcuoglu on Google's blog [1].

Google also reports safety changes against Argon. It says successful cheating on Terminal-Bench 4.0 fell to 0.0%, high-severity misalignment alerts in live coding dropped by almost 80% and unsupported claims fell by about half [1]. Prompt-injection attack success on the Gray Swan benchmark is 0.0% against 0.7% for Argon [1]. These are Google's internal results.

Under the hood: what Google discloses

Google has not stated a context window, a parameter count or an architecture for Carbon [1][2]. It says the model builds on Argon's safety stack, which covers training, misuse defenses, prompt-injection resilience and chain-of-thought monitoring [1]. Google also says it is taking part in a voluntary US government pre-release testing process [1]. For another recent model release with fuller specs, see Claude Haiku 5.5 Launch: $0.10 Input, 1M Context, Benchmarks.

Benchmarks: Google-reported and independent

Google lists 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching [1]. It reports 1768 Elo on WebDev Arena, fourth place and up 89 points [1]. Google says internal tests show more than twice Argon's success rate in generating secure code, and that the Chrome team saw over 10 times better patch quality than another lab's large model, without naming it [1].

The independent picture is more modest than the headline. Artificial Analysis puts Carbon first overall by one point, tops 4 of 9 evaluations and does not lead on coding or knowledge work, and the coding agent index puts it three points behind Opus 5.5 [2]. A one-point lead could fall within run-to-run variance [2].

Pricing, availability and how to access it

The introductory price is $2 per million input tokens and $10 per million output tokens, the same as Argon, with cached input discounted by 95% [1][2]. After the introductory period the price rises to $4 and $20, and no end date has been published [1][2]. Access starts this week with a subset of cyber defenders and then extends to more trusted partners through Google's Fairwind Program [1]. Developers, businesses and consumers come later, as soon as possible and subject to more guardrail work, with no firm date [1].

Limitations, trade-offs and alternatives

Artificial Analysis reports that Carbon's hallucination rate rose from 4% for Argon to 11% on its AA-Omniscience test, although accuracy also improved [2]. Most capability figures beyond the leaderboard come from Google's own announcement, and Carbon is not publicly available, so outside teams cannot yet repeat the results [1][2].

Most readers cannot use Carbon today. Developers who need a strong coding agent now can compare Claude Opus 5.5, which leads Carbon on the Coding Agent Index, and Argon at the same introductory price, and revisit Carbon when general access opens [2]. For open-weight alternatives, see Mistral Large 4 Preview: 1.05T Open-Weight MoE, Specs and Pricing.

Sources

1. Google. "Gemini 4 Carbon: our next frontier model," Google blog, October 28, 2026. blog.google
2. ExplainX. "Gemini 4 Carbon: Index 59, $2/$10 Price, Who Gets It." explainx.ai

Read More on Depth Grid

• Nvidia in Talks to Deepen Stake in or Buy Reflection AI
• Claude Haiku 5.5 Launch: $0.10 Input, 1M Context, Benchmarks
• Meta Muse Wonder Launch: 1M Context, $4 API Price, Benchmarks
• Mistral Large 4 Preview: 1.05T Open-Weight MoE, Specs and Pricing

Article by Mahesh | Depth Grid