Google’s strongest model is about to cost twice as much to run. Gemini 4 Carbon’s API rates double on December 1, rising from $2 to $4 per million input tokens and from $10 to $20 per million output tokens, when introductory pricing ends on November 30.
The New Rates
The standard tier is where most developers will feel it. Input moves from $2 to $4 per million tokens, output from $10 to $20, and cached input from $0.10 to $0.20. A workload costing $100 in November costs $200 for the same tokens in December.
The cheaper tiers double too. Batch and Flex input rises from $1 to $2 per million tokens, with output up from $5 to $10, so the 50 percent batch discount survives but the bill still doubles. Priority input climbs from $3.60 to $7.20, and its output from $18 to $36.
Carbon has no free tier, and Google counts every token it processes. Thinking tokens bill at the output rate, as do the internal tool instructions and function declarations added when grounding is used. A model that reasons at length therefore costs more than its headline rates suggest.
Why the Bill Can Grow Faster
Carbon is built to spend tokens. Google says it uses more of them than Argon to explore thoroughly, reason deeply and run verification loops, which is why it leads on hard agentic work. The same task can therefore burn far more output tokens on Carbon than on its cheaper sibling.
Google’s own case is that Carbon still earns its keep. It ranks first on CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench at more than twice lower cost per task than the next best outside model, even after the price rise. That claim rests on Google’s own measurements, not independent tests.
Argon, Google’s default for most coding and knowledge work, carries exactly the same rates and doubles on the same day. The choice between the two models is about token appetite, not price per token. For lighter work, the cheaper Gemini 4 Neon, launched last week, sits below Carbon.
Our Read
This is a scheduled step-up, not a surprise hike. Google printed the December rates in its launch footnote on November 10, giving teams three weeks to prepare. The intro price was always a trial rate, and December is when the trial ends.
The practical move is to finish one-off batch work before November 30, when the same tokens cost half as much. After that, Batch API work still halves the bill, cached input stays far cheaper than fresh input, and only the hardest tasks should reach Carbon at all.
Why It Matters
For teams already running Carbon in production, the December bill doubles with no change in usage. Pilots signed off on November’s numbers will be wrong by a factor of two, and the ten-day window is the last chance to re-baseline budgets or shift lighter work to Neon or Argon.
Background
Carbon first appeared on October 28 for vetted cybersecurity defenders, then opened to all paying developers on November 10, with AI Ultra subscribers getting it in the Gemini app, AI Mode and Antigravity. Google calls it its strongest model, built for long-horizon agentic work, deep research and cyber defense.
Source: Google DeepMind’s Carbon launch post and Google’s Gemini API pricing page


0 Comments