Google has just taken the wraps off its most powerful AI model yet — for everyone. Gemini 4 Carbon, which until now was limited to a small group of cybersecurity experts, is now open to developers, businesses, researchers and paying consumers. If you have been waiting for Google's heavyweight model to escape the lab, the wait is over.
What is Gemini 4 Carbon?
Google DeepMind calls Carbon its strongest model for cybersecurity and long-horizon agentic workflows — the jobs that take many steps, repeated checks and extended reasoning. Think complex coding, debugging, scientific research, analysing long documents and hunting for software vulnerabilities. Its sibling, Gemini 4 Argon, stays the default for most everyday coding and knowledge work; Carbon is the one you reach for when the job is genuinely hard and long.
The Benchmarks
Carbon arrives with strong numbers behind it. Google says it ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench — and at more than twice the lower cost per task than the next-best external model. Cursor puts Carbon at 60.4% on CursorBench at high effort. For balance: on one agentic benchmark, Anthropic's Fable 5.2 still edges it out (57% to Carbon's 54%), though Google says Carbon does it at a far lower cost per task.
Price — and the December Deadline
For developers, Carbon costs $2 per million input tokens and $10 per million output tokens — but only until November 30. From December 1, the price doubles to $4 and $20. Google is also offering a 95% discount on cached input tokens during the introductory period. In other words, the cheap window is open right now, and it closes in under three weeks.
Where You Can Use It
Carbon is available through the Gemini API in Google AI Studio, Gemini for business, and Android Studio. Google AI Ultra subscribers get it in the Gemini app, AI Mode in Search and Google Antigravity. Eligible scientists and researchers can apply for six months of free access through Google's research programme, and cybersecurity professionals can apply through the Fairwind programme. Google is also previewing "/swarm" in Antigravity: give one instruction and several AI agents split the work between them — one plans, one builds, one checks.
Why It Matters
Carbon's public opening lands in a crowded week for Google — Gemini Robotics 2.5 just launched with a factory pilot, and DeepMind's AlphaProof Code formally verified parts of the Signal messaging protocol. The pattern is clear: Google is moving its most capable models out of research demos and into real, long-running work. For anyone doing serious coding or security research, Carbon is worth a test run now, while the introductory price lasts. Just remember the December 1 deadline — and that Argon is still the sensible default for everyday jobs.
Comments
Post a Comment