Gemini price increase, December 2026: what it means for your bill

When Google announced Gemini 4 Argon on 30 September 2026, it described the launch price as introductory and said in a footnote that $4 per million input tokens and $20 per million output tokens would apply afterwards. Google's API pricing page now gives the date: the higher prices start on 1 December 2026. If you moved work to Argon at the launch price, your December bill for that work will double unless you change something. This page sets out the new prices, what they do to a typical bill and what you can do in the five weeks before the change. All prices are list prices in US dollars per million tokens, checked on Google's pricing page on 25 October 2026.

What is changing

Every Argon rate doubles: standard, batch, Flex and Priority, for input, output and cached input. Output includes thinking tokens, which Argon bills at the output rate.

Two other details matter for cost. Argon is not available on the Gemini API free tier. And Google's migration notes say Argon does not support explicit context caching: caching is implicit, on by default, for prompts of at least 4,096 tokens. Grounding with Google Search or Maps is billed separately, and Google lists no change to it: the first 5,000 a month are free, shared across Gemini 3 and later models, then $14 per 1,000.

What it does to a real bill

Take the default workload in our API pricing calculator: 50,000 requests a month, 6,000 input tokens and 400 output tokens per request, with 60% of the input served from cache.

• Until 30 November: 2,400 uncached input tokens at $2, 3,600 cached tokens at $0.10 and 400 output tokens at $10 come to about $0.0092 per request, or $458 a month.
• From 1 December: the same traffic costs $916 a month.
• From 1 December on Batch or Flex: $458 a month, the same as standard processing today.

Your own figure depends on how many thinking tokens your requests generate. Because thinking is billed as output, a request with a short visible answer can still carry a large output charge, and that part of the bill doubles too.

How Argon compares after the change

Here is the same example workload on other current models, at list prices on each provider's own page. Cache write charges on Anthropic models are not included.

After 1 December, Argon's list rates match Claude Opus 5.5 and remain well below GPT-6 Astra. So the increase does not make Argon expensive among flagship models; it removes the launch discount. The question for most teams is not whether to leave Argon, but which of their requests actually need a flagship model at the new price.

Six things to do before 1 December

1. Find out how much you spend on Argon

Export your usage and look at spend by model. If Argon is a small share, the increase is a line item; if it is most of the bill, it needs a plan. Our free Bill Analyzer does this on your own computer.

2. Move work that can wait to Batch or Flex

Both are half the standard price, so after the increase they cost what standard costs now. Nightly summaries, document processing, evaluations and backfills are the usual candidates.

3. Lower the thinking level where the task is simple

Argon offers low, medium and high thinking levels, with medium as the default. Thinking tokens are billed as output. For classification, extraction and short factual answers, test the low level on your own examples; if quality holds, you pay for fewer output tokens.

4. Make sure implicit caching is working

Caching on Argon is automatic for prompts of 4,096 tokens or more, and cached input costs 95% less than uncached input. It only works when the start of the prompt is identical between requests. Put instructions, examples and fixed documents first and anything that changes, such as dates, user names or the question, last. Check the cached token count in your usage data rather than assuming.

5. Test a cheaper model for the simple share of traffic

Split each feature into simple and complex requests. At December prices, Claude Sonnet 5.5 and GPT-6.1 Sol list at half Argon's rates, and Gemini 3.8 Flash lower still until the end of the year. Route only the requests where your tests show quality holds.

6. Update your budget and your own prices

If you sell an AI feature on a fixed plan, your cost per user on Argon doubles in December. Re-run your margin with the new rates now, while there is time to change limits or plans.

Gemini 3.8 Flash in January

Gemini 3.8 Flash has its own increase one month later. Google's pricing page says the current prices apply through 31 December 2026 and double from 1 January 2027: input from $0.75 to $1.50, output from $3.75 to $7.50, cached input from $0.075 to $0.15 and cache storage from $0.50 to $1.00 per million tokens per hour. On our example workload that is $179 a month today and $357 from January. If you plan to move Argon traffic to 3.8 Flash, budget on the January price. Google lists no announced changes for Gemini 3.5 Flash-Lite or 3.1 Flash-Lite as of 25 October 2026.

Check your own exposure

Open the API pricing calculator, enter your own request volume and token counts, and tick "Use announced future prices" to see December and January costs side by side with every other model. For a full review of your workload, including which requests can safely move to a cheaper model, see the LLM Cost Report.