Gemini 4 Argon and Carbon API Prices Double to $4/$20 on 1 December

Correction, 13 November 2026: an earlier version of this article said Gemini 4 Carbon was still limited to cyber defenders and had no public date. Google's changelog lists gemini-4-carbon as generally available on the Gemini API from 10 November 2026. The headline, table and advice below have been corrected.

Google's Gemini API pricing page now shows two prices for Gemini 4 Argon: $2 per million input tokens and $10 per million output tokens through 30 November 2026, then $4 and $20 from 1 December. Argon was announced on 30 September for a small group of cyber defenders, and Google opened it to developers on 13 October on the paid tier of the Gemini API. A second model, Gemini 4 Carbon, was announced on 28 October for cyber defenders and went generally available on 10 November at the same two prices and with the same 1 December increase. If you adopted either model this autumn, your bill doubles on 1 December.

What happened

Gemini 4 is Google's first new major model generation since Gemini 3, and it shipped in phases. In the 30 September post, Google DeepMind's Koray Kavukcuoglu said Argon would go first to trusted cyber defenders through the Fairwind Program and to the US government's voluntary pre-release testing process, with developers, enterprises and consumers after that. The post put the reason plainly: "Safely releasing frontier capabilities at this level requires a phased approach."

Developer access came on 13 October. Google's developer post lists the paid tier of the Gemini API in Google AI Studio and Android Studio, and access in AI Studio and Google Antigravity for Google AI Ultra subscribers. Android Headlines reported the consumer rollout to the Gemini app and AI Mode in Search for AI Pro and AI Ultra subscribers on 14 October. Google's October recap confirms Argon is now available across AI Studio, the Gemini API, Google Cloud, Search and the Gemini app.

Gemini 4 Carbon was announced on 28 October as Google's strongest model, aimed at long-horizon engineering and cyber defence. At first it went only to a subset of cyber defenders through Fairwind, with broader access promised "as soon as possible" and no date. Thirteen days later, on 10 November, Google's changelog listed gemini-4-carbon as generally available on the paid tier of the Gemini API. On 12 November the same changelog added a third model, Gemini 4 Hydrogen, which is generally available with a free tier and a paid price of $1.50 in and $7.50 out per million tokens.

The numbers

For comparison, the same pricing page lists Gemini 3.1 Pro Preview at $2 input and $12 output per million tokens for prompts up to 200K tokens. After 1 December, Argon costs twice as much per input token as Google's previous Pro model.

Why it matters

Introductory pricing is a familiar SaaS move, and this cliff is close. Teams that ran October evaluations at $2 and $10 and then sized budgets on that number will be off by a factor of two from December. Google's pricing page also says that every token the model processes is billed, including thinking tokens. With an output limit of one million tokens, a single runaway request can now be expensive in a way it could not be at 64K.

Carbon matters for a different reason. Google did what Anthropic does with Mythos and what OpenAI does with its cyber variants: it shipped its strongest model to vetted defenders first. Google's own Carbon numbers are cyber-heavy: 88.2% on SEC-bench Pro for finding vulnerabilities and 54.8% on ExploitGym. The difference is how long the gate lasted. Thirteen days after the announcement Carbon was generally available on the paid API, so the best Google model is now something you can build on, at a price that doubles on 1 December.

What sceptics say

Every benchmark in Google's posts is Google-reported, and several are described only as "leading" with no score. Independent comparisons are thin, and the first one from a rival is not flattering. In Anthropic's Fable 5.2 launch table, Gemini 4 Carbon scores 41.1% on FrontierCode 2.0 (at medium effort) against 49.1% for Claude Fable 5.2, though Carbon wins AutomationBench in the same table, 53.9% to 51.4%. Rival vendors pick their own benchmarks too, so neither table settles anything.

The phased release also had a cost. On 24 September 9to5Google reported that Kavukcuoglu, speaking at an event run by The Information, said Google wanted an early post-training version out "as soon as possible". That turned into a cyber-defender preview, then a paid-only developer launch, then a second model that stayed gated for thirteen days before general release. For anyone planning a migration, the honest summary is that Google's roadmap dates have been soft this year.

What it means for engineering and DevOps teams

Here is how I would handle it if Argon were in my stack today.

• Re-run the budget at December prices now. Use $4 and $20, not $2 and $10. If the number doesn't work, decide before 1 December, not after the first invoice.
• Move offline work to Batch or Flex. From December, both are half the standard price at $2 and $10. Nightly evals, document processing and backfills do not need standard-tier latency. My guide to cutting LLM API costs covers the batching pattern.
• Cap output tokens explicitly. A 1M-token output limit is a ceiling, not a target. Set a per-request maximum and alert on outliers.
• Budget Carbon at December prices too. It has been generally available since 10 November as gemini-4-carbon, at the same introductory $2 and $10, and it also moves to $4 and $20 on 1 December. It has no free tier.
• Keep a cheaper model or a rival as fallback. Gemini 4 Hydrogen, generally available since 12 November at $1.50 and $7.50, is the low-cost Gemini 4 option, and Gemini 3.1 Pro is still listed. A gateway that can route by model and price, like the ones in AI Gateways in 2026, turns a price change into a config change.

# Rough monthly cost for Gemini 4 Argon before and after 1 Dec 2026.
# Prices per 1M tokens from the Gemini API pricing page (standard tier).
PRICES = {"to_nov30": (2.00, 10.00), "from_dec1": (4.00, 20.00)}

def monthly_cost(input_mtok: float, output_mtok: float, period: str) -> float:
    p_in, p_out = PRICES[period]
    return input_mtok * p_in + output_mtok * p_out  # output includes thinking tokens

for period in PRICES:
    print(period, round(monthly_cost(500, 80, period), 2))  # 500M in, 80M out

FAQ

How much does Gemini 4 Argon cost?

Through 30 November 2026, $2 per million input tokens and $10 per million output tokens on the standard tier. From 1 December, $4 and $20. Batch and Flex are half the standard price, and there is no free tier.

Can I use Gemini 4 Carbon?

Yes. Since 10 November 2026 it is generally available on the paid tier of the Gemini API as gemini-4-carbon. Before that it was limited to Google's Fairwind Program for cyber defenders. It is not on the free tier.

Is Gemini 4 Argon better than Gemini 3.1 Pro?

On Google's benchmarks, yes, but it costs more. After 1 December, Argon's input price is double Gemini 3.1 Pro Preview's $2 for prompts up to 200K tokens. Test both on your own tasks before switching.

Related: How to Set Up LiteLLM Proxy: Budgets, Fallbacks and Virtual Keys (2026), Self-Hosting LLMs vs API Costs in 2026: The 62% Utilisation Break-Even and AI Found the Bugs: 23,000 Findings in a Month, a 3-Day Patch Mandate, and the End of curl's Bounty.