Google DeepMind opened Gemini 4 Carbon to developers and paid users on November 10, 2026. On Google's official pricing page, Carbon (gemini-4-carbon) and Argon (gemini-4-argon) have identical paid-tier prices, and both rise on the same day. Neither is available on the Free tier.
- Through November 30, 2026: $2 per 1M input tokens and $10 per 1M output tokens (output includes thinking tokens).
- From December 1, 2026: $4 input and $20 output per 1M tokens.
Because list prices are the same, the choice comes down to token usage and success rate on your own tasks, not price or leaderboard rank. Google states that Argon has lower token spend per turn than Carbon across complex coding, agentic, math, and science tasks; DeepMind says Carbon targets harder multi-step coding, debugging, and open-ended research and uses more tokens than Argon. These are official claims; we have not tested either model.
Simple formula: cost per call ≈ input tokens/1M × input price + output tokens/1M × output price (count thinking tokens as output). Cost per successful task = total cost of the batch, including failures and retries ÷ number of successes.
How to compare: run the same batch of real, de-identified tasks on both models with identical prompts, tools, and acceptance criteria; log pass/fail, input tokens, output tokens including thinking, and billed amount; compute success rate and cost per successful task at both pre- and post-December 1 prices; retest before making long-term decisions.
Sources: Google Gemini API pricing · DeepMind blog · Carbon model page · Argon model page
Reviewed November 11, 2026, Asia/Shanghai.