Osegi
AI·2 min read

Google Opens Gemini 4 Carbon to Developers and Businesses

Google has widened access to Gemini 4 Carbon, its model for long, multi-step work in coding, science and security, with launch pricing of $2 and $10 per million tokens.

Google Gemini 4 Carbon logo displayed above abstract code and data visuals representing the AI model's launch.
Image: Webrazzi

Google has rolled out Gemini 4 Carbon to a broader audience, weeks after introducing it with limited access at the end of October. The company says the model is built for long, multi-step tasks such as complex information processing, software development, scientific research and cybersecurity.

Within the Gemini 4 family, Argon is positioned for most software development and everyday knowledge work. Carbon targets harder jobs: deeper reasoning, multi-step coding and debugging, and open-ended research. According to Google, it uses more tokens than Argon to investigate a problem thoroughly and can run long verification loops.

Availability and pricing

Paid developers can reach Carbon through the Gemini API in Google AI Studio, and companies through Gemini for business. Google AI Ultra subscribers get it in the Gemini app, AI Mode and Antigravity, Google's agent development environment.

Carbon shares Argon's launch pricing: $2 per million input tokens and $10 per million output tokens until November 30, 2026. After that the rates are due to rise to $4 and $20. Anthropic's Claude Opus 5.5 is priced at $4 and $20, while Claude Fable 5.2 and Mythos 5.2 cost $8 and $40, so Carbon's promotional price is a quarter of Fable 5.2's. Token price is not the whole story, though: the number of tokens a model needs, retries and human intervention all shape total cost.

Agents, coding and security

Google is highlighting agent workflows. A /swarm feature, in preview in Antigravity, splits one task across several AI agents. Carbon handles architecture, coordination and oversight, while faster models work in parallel on other parts of the project.

Google says Carbon ranked first on CursorBench 4.0, Perplexity WANDR and Zapier AutomationBench, at under half the per-task cost of the next external model. On the Multimodal Knowledge Agent test built with Artificial Analysis, it placed second at an average of six times lower cost per task than the leader.

In Project Fortify, Google's internal tests, a special Carbon version offered through the Fairwind Program reached 100% recall in vulnerability scanning and 95% precision in classification. Google also says it helped uncover previously unknown flaws in some open-source software. The company expanded its researcher program to include Carbon: accepted scientists get six months of free access to Argon and Carbon on Antigravity's highest usage tier.

Apps worth a look

SUMO: the number puzzle

Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.

Tags:googlegemini 4 carbongemini apiantigravityai pricinganthropic

More from this day

See all →

More in: AI

See all →