Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)

Gemini 4 Carbon Takes First Place on Artificial Analysis Leaderboard with 59 Points at Argon's Price

Google has unveiled Gemini 4 Carbon, the successor to Argon, aimed at long-running AI agent tasks and code security. It takes first place in the Artificial Analysis composite leaderboard with 59 points, and the price stays the same: $2 per million input tokens and $10 per million output tokens. Access is currently limited to a narrow group of cyber defenders, and there is no date for a broad launch. That is why it makes sense to establish a baseline on Argon now.

AI-processed from TG: @neurogen_news (Neurogen); edited by Hamidun News
Gemini 4 Carbon Takes First Place on Artificial Analysis Leaderboard with 59 Points at Argon's Price
Source: TG: @neurogen_news (Neurogen). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @neurogen_news (Neurogen) ↗
Editor-in-Chief
Cached input tokens are 95% cheaper$0.10 per million
Key metric #2Cyber defenders get access first, then partners through the Fairwind program
Key metric #3Google is undergoing a voluntary pre-release review by the US government

Google has unveiled Gemini 4 Carbon, a new flagship model that follows Argon. It is built for two tasks: long agentic workflows, where the AI carries out a chain of actions on its own, and code security. On the updated Artificial Analysis leaderboard (an independent composite table that compares models on a set of tests), Carbon takes first place with 59 points. It costs the same as Argon. Not everyone will be able to try it, however, and not right away.

What Google claims

According to the company, Carbon delivers the best result among models in cybersecurity. It also tops the updated Artificial Analysis index. 59 points is a composite score across a set of tests. As a benchmark the number is useful, but it does not replace testing on your own tasks.

A separate claim concerns secure code. If Carbon really is roughly twice as good as Argon at writing and reviewing secure code, the difference will show up in everyday work, for example when reviewing pull requests or searching for vulnerabilities in a large repository. For now, this is Google's own claim. The source contains no independent measurements.

The price did not go up

Normally, moving to a stronger model means a higher rate. Here that does not happen: Carbon costs the same as Argon.

  • $2 per million input tokens;
  • $10 per million output tokens;
  • cached input tokens are 95% cheaper, that is, $0.10 per million.

The cache discount matters especially for agents. Such systems repeatedly send the model the same long context: instructions, pieces of the codebase, the history of steps. If that part of the request is cached, the main input bill drops by almost a factor of twenty. For long agentic tasks, that is a meaningful saving.

Who gets access and when

This is where the restrictions begin. Starting this week, Carbon is available to a narrow group of cyber defenders. The model will then open to trusted partners through the Fairwind program. For other developers, businesses, and ordinary users, Google promises a release "as soon as possible" but does not name a date.

The logic of caution is clear. A model that is good at finding and fixing vulnerabilities could, in theory, also help those who exploit them. That is why it goes first to those who defend. In parallel, Google is undergoing a voluntary pre-release review by the US government. This is neither a ban nor a license, but it is a signal: major labs increasingly coordinate powerful models before public launch.

What to do in practice

Argon is available now at the same prices, and that is where you should establish a baseline. Assemble a small set of real tasks: bug fixing, vulnerability analysis, long agentic workflows. Run it now and save the results. When Carbon opens up, you can compare both models on your own data right away and see whether switching is worth it. If the claimed advantage in secure code holds up, the switch will not add a single dollar to your bill.

If you work in security and want early access, it is worth asking Google or its partners about the terms of Fairwind. Everyone else will have to wait and prepare their tests. For now, Carbon has figures from Google itself and first place on Artificial Analysis. There are no independent open verifications and no timeline for a broad launch, so it is too early to draw final conclusions.

Story

Key Questions Answered

What Google claims?

According to the company, Carbon delivers the best result among models in cybersecurity. It also tops the updated Artificial Analysis index. 59 points is a composite score across a set of tests. As a benchmark the number is useful, but it does not replace testing on your own tasks. A separate claim concerns secure code.

What are the key details on The price did not go up?

Normally, moving to a stronger model means a higher rate. Here that does not happen: Carbon costs the same as Argon. $2 per million input tokens; $10 per million output tokens; cached input tokens are 95% cheaper, that is, $0.10 per million. The cache discount matters especially for agents. Such systems repeatedly send the model the same long context: instructions, pieces of the codebase, the history of steps. If that part of the request is cached, the main input bill drops by almost a factor of twenty.

Who gets access and when?

This is where the restrictions begin. Starting this week, Carbon is available to a narrow group of cyber defenders. The model will then open to trusted partners through the Fairwind program. For other developers, businesses, and ordinary users, Google promises a release "as soon as possible" but does not name a date. The logic of caution is clear. A model that is good at finding and fixing vulnerabilities could, in theory, also help those who exploit them. That is why it goes first to those who defend.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…