Skip to content
Rush Commerce
AI & Automation3 min read

Gemini 4 Carbon hits the paid API: $2/$10 until Nov 30

Gemini 4 Carbon is open to paid Gemini API developers at $2/$10 per million tokens, doubling December 1. Google says Argon still fits most work. Route only the hard tasks.

Two weeks ago we told you not to plan around a model you could not call. Now you can call it. Gemini 4 Carbon, Google's top model, opened to every paying Gemini API developer today. The model ID is gemini-4-carbon. The introductory price is $2 per million input tokens and $10 per million output. On December 1 that doubles. And Google's own launch post says most work should still go to Argon.

What actually happened

Per Google's announcement, dated November 10:

  • Access: the Gemini API paid tier in AI Studio and Android Studio, plus Gemini for business. Google AI Ultra subscribers get it in the Gemini app. Before today, only cyber defenders in Google's Fairwind Program had it.
  • No free tier: the Gemini API pricing page lists Carbon as "Not available" on the free tier.
  • Price: $2 input / $10 output per million tokens through November 30, then $4 / $20. Thinking tokens bill as output. Cached input is $0.10, then $0.20, and the pricing page says caching is implicit only. Batch and Flex run at half price.
  • Google's own guidance: Argon is recommended for most software engineering and knowledge work. Carbon is for harder tasks and uses more tokens than Argon.
  • Preview: a /swarm command in Google Antigravity splits one prompt across a team of agents, with Carbon taking the planning and audit roles.

VentureBeat's review adds context. Carbon scores 60 on the Artificial Analysis Intelligence Index, behind Claude Fable 5.2 at 65. It beats Fable on Zapier's AutomationBench at less than half the cost per task. It trails Fable on two coding benchmarks. VentureBeat also notes that Google has not given Vertex AI availability or a country-by-country rollout.

Why Gemini 4 Carbon pricing matters for your business

The $2 price is a three-week coupon. If you budget a workflow at $2/$10, you will be wrong on December 1. Do the math at $4/$20. That is the real price.

Per-token price is not per-task price. Carbon thinks longer than Argon. At the same list price, more tokens per answer means a bigger bill per answer. Measure cost per finished task on your own work, not cost per million tokens on a pricing page.

This is a routing decision, not an upgrade. The right setup for most small businesses is a cheap model for volume, a mid-tier default, and a top model only for the jobs that fail without it. Carbon is a candidate for that top slot. Run your existing eval set against it this week, while the trial is cheap. Promote it only where the pass rate earns the price.

Key takeaways

  • Gemini 4 Carbon is open to paid Gemini API developers as of November 10; no free tier
  • $2/$10 per million tokens through November 30, then $4/$20; thinking bills as output
  • Google itself recommends Argon for most work; Carbon uses more tokens
  • Vertex AI availability and regions are not yet published
  • Eval it now at the intro rate, budget at $4/$20, route only the hard tasks

The best model is the cheapest one that passes your test. We build model routers that send each job to the right tier and track cost per task, so a price change is a config edit. See how we build AI systems.

Sources: Google: Gemini 4 Carbon now available, Gemini API pricing, VentureBeat.

  • #gemini-4-carbon
  • #google
  • #api-pricing
  • #model-routing
  • #ai-agents
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.

Keep reading

AI & Automation3 min read

Skyfall's Steve-001 AI agents run 3 businesses: one up 127%

Skyfall AI's Steve-001 agents run three acquired businesses. One doubled revenue, one is down 3%. The lesson for owners: write your approval tiers first.

Read it
AI & Automation3 min read

NEAR AI breach: unpatched SGLang bug exposed agents and keys

NEAR AI says an unpatched SGLang RCE let an attacker reach 100 user accounts, 18 agents and two API keys in logs. Patch your inference stack and scrub logs.

Read it