Gemini 4 Carbon’s Intro Price Ends November 30

Google made Gemini 4 Carbon available to paying API users on November 10, 2026, after an earlier limited release to cybersecurity defenders. Its Standard API price is $2 per million input tokens and $10 per million output tokens, including thinking, through November 30. On December 1, those rates rise to $4 and $20. That makes the date of a production rollout part of the budget decision, not a footnote. Carbon has no Free API tier. Google’s launch announcement describes the access change; its API price list gives the dated rates.

What changed on November 10

Google says Carbon had been available to a selected group of cybersecurity defenders through its Fairwind Program. The November 10 announcement expands access through Google AI Studio’s API and Gemini for business for paying developers and enterprises. The same announcement says Carbon is rolling out to Google AI Ultra subscribers in other Google products. An Ultra subscription and a paid API account are different ways to access the model; do not assume an app subscription makes API calls free. For a developer planning an integration, the relevant starting point is the paid API price and the model’s availability in the intended account. Google’s announcement states these access routes, while the pricing page lists the API Free tier as unavailable.

Google positions Carbon for difficult, multi-step coding, debugging and research. It says Gemini 4 Argon remains the choice for most software engineering and knowledge work. Those are the maker’s descriptions, not results from our tests. The choice depends on whether Carbon improves the outcome of your own difficult tasks enough to justify its total billed tokens. The site’s model tracker can help you keep the release in view; use Google’s current docs for the charge you will actually pay.

The Standard price changes on December 1

Google’s Gemini Developer API pricing page lists the following paid Standard rates in US dollars per million tokens. The output column includes thinking tokens, so a visible answer alone does not describe all billed output. Carbon has no Free API tier.

Standard API meter	Through November 30, 2026	From December 1, 2026
Uncached input, per million tokens	$2.00	$4.00
Output including thinking, per million tokens	$10.00	$20.00
Cached input, per million tokens	$0.10	$0.20

Each listed token rate doubles on December 1. Google specifies that Carbon’s cached-input rate applies to implicit caching; explicit caching is not supported for this model. Do not put every repeated prompt in the cached column unless your actual usage shows a cache hit. The pricing page also has separate Batch, Flex and Priority tabs. Their rates and operating conditions are distinct from the Standard rates above. Any comparison must hold the service tier constant.

The same official pricing page lists Argon at the same dated Standard input and output rates. That does not make Carbon and Argon equal in cost for the same task. Google says Carbon may use more tokens on complex work to reason and verify. If Carbon uses more billed output tokens, identical per-token prices still produce a larger bill. The site’s API price comparison explains why input and output must be accounted for separately; its older comparison is not a substitute for this dated Carbon rate card.

Price the same workload on both sides of the deadline

Here is an illustrative, fixed token budget, not a measured Carbon run: suppose a set of calls uses 1 million uncached input tokens and 200,000 output tokens, including thinking. Assume every call succeeds, no input is cached, and there are no grounding or other tool charges, retries, taxes or separate platform fees.

When those tokens are billed	Input calculation	Output calculation	Standard token subtotal
Through November 30	1 × $2 = $2	0.2 × $10 = $2	$4
From December 1	1 × $4 = $4	0.2 × $20 = $4	$8

The formula is (input tokens ÷ 1,000,000 × input rate) + (output tokens ÷ 1,000,000 × output rate). Holding the token counts fixed isolates the published price change: this example rises by $4, or 100%. It does not predict the bill for a real prompt, because the actual input, thinking and answer tokens can differ. For a workload that runs across both dates, split actual usage by billing date and apply the corresponding rates. A November pilot at the introductory rate does not lock that rate into December.

Keep other meters separate. Google’s price page lists charges for some grounding usage and separate processing tiers. If your workflow invokes those features, include the applicable line items from the account’s usage record. The site’s API cost calculator is useful for token arithmetic, but the dated figures and any cache hits still need to come from the current maker price page and your own usage. For how repeat requests can affect the input side of a bill, see prompt caching explained.

Decide on cost per successful task

Start with one representative difficult task for which Argon’s result is not good enough. Write a pass rule before comparing models: for example, whether the code meets specified tests or a research answer cites the required primary documents. Run the same task and evaluation rule for both models under the same service tier. Record billed input tokens, billed output and thinking tokens, retries and any tool charges for each attempt. Then calculate total billed spend ÷ accepted results. If no result passes, cost per accepted result is undefined; report the failure rather than a misleading low price.

This procedure is an editorial recommendation, not a claim that we ran the models. Google reports favorable Carbon benchmark results in its announcement, but a vendor-reported benchmark cannot establish the price or quality of your own workload. Carbon may be worthwhile when its extra work produces a materially better accepted result; Argon may remain the cheaper choice for routine tasks. Use a small, defined comparison before changing a large workload, and rerun the budget with December rates if the deployment will continue past November. The site’s evaluation cost guide shows how to keep generation and judgment costs separate when you evaluate those results.

Sources

• Google: Gemini 4 Carbon is now available — November 10 announcement, access change and maker’s model positioning.
• Google: Gemini Developer API pricing — Carbon and Argon Standard rates, date transition, Free tier and caching conditions.

All pages read on November 12, 2026.

Free toolWhat will 10,000 chats a month cost?Estimate my bill