Gemini 4 Carbon is now live for paying developers and Google AI Ultra subscribers — $100/month in the US, ₹8,499/month in India, two weeks after its initial restricted release to cybersecurity defenders. It’s not a standalone product. It’s a model tier, accessible via API or inside existing tools: the Gemini app, AI Mode in Search, and Antigravity, Google’s coding assistant.
No new hardware. No new app. Just more compute, more latency, and more tokens burned per task. Google says it costs the same per token as Gemini 4 Argon at launch: $2 per million input tokens, $10 per million output tokens.
That jumps to $4 and $20 after an introductory period. But because Carbon thinks longer and deeper, especially on complex debugging, scientific inference, or multi-step reasoning, the same request can cost more than with Argon. Argon remains Google’s recommendation for everyday coding and office work. Carbon is for the outliers: a researcher running simulations across 200 variables, a dev tracing a memory leak across five microservices, a team stress-testing a zero-trust architecture before rollout.
It’s also the brain behind /swarm, a new Antigravity feature previewed today. One prompt spawns a team of AI agents: Carbon leads, cheaper models handle boilerplate, others verify outputs. Google says Carbon topped the Artificial Analysis Intelligence Index on first appearance, though its hallucination rate spiked on one knowledge test.
It leads three coding and automation benchmarks (Cursor, Perplexity, Zapier) at less than half the cost per task of the next best model. Its offensive security capabilities remain locked. The cyber-specific version stays exclusive to vetted defenders under Google’s Fairwind programme. The public Carbon blocked 99% of attack attempts in internal red-team tests, but that’s all Google is saying about its defensive chops right now.
Developers can start using it today via the Gemini API. AI Ultra subscribers get access in-app immediately. No global rollout date was given, just ‘rolling out’. And no price was announced for enterprise contracts beyond the per-token API rates.
The next step? Wider availability. Or tighter controls. Google hasn’t said which.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.