Google Gemini API Pricing for Argon: What a Month Costs Before and After December 1
A workload of 30,000 requests a month (assumptions below) costs $480 on Gemini 4 Argon in November and $960 in December, and nothing in the code changed. Google made the model broadly available on October 13, 2026 at $2 per million input tokens and $10 per million output tokens through November 30. From December 1 the rates are $4 and $20. The figures here are as of October 15, 2026, from the Gemini API pricing page and VentureBeat's report of the launch.
The rates
USD per million tokens, paid tier, from the Gemini API pricing page. The page says Argon is not available on the free tier.
| Item | Through Nov 30, 2026 | From Dec 1, 2026 |
|---|---|---|
| Standard input | $2.00 | $4.00 |
| Standard output, thinking tokens included | $10.00 | $20.00 |
| Context caching (implicit only) | $0.10 | $0.20 |
| Batch input | $1.00 | $2.00 |
| Batch output, thinking tokens included | $5.00 | $10.00 |
Source: Gemini API pricing page, checked October 15, 2026.
The words "thinking tokens included" carry the budget. The model's reasoning is billed at the output rate, and you do not see it in the reply text. The page lists no price tiers by prompt length for Argon, which keeps the arithmetic simple.
The workload
These are assumptions, not measurements. Swap in your own.
- 30,000 requests a month (1,000 a day).
- 3,000 input tokens per request, 90 million a month.
- 400 tokens of visible answer plus 600 tokens of thinking, so 1,000 billed output tokens per request, 30 million a month.
Step by step
Using the rates from the pricing page above, input: 90 million tokens x $2 per million = $180 now, x $4 = $360 from December.
Output: 30 million tokens x $10 per million = $300 now, x $20 = $600 from December.
Total with no caching: $480 in November, $960 in December.
Now let half of the input hit the cache: 45 million tokens at the cached rate, 45 million at the full rate. The page says only implicit caching is offered, so you cannot force a hit, and half is a guess.
| Scenario | Nov 2026 | Dec 2026 |
|---|---|---|
| Standard, no cache hits | $480.00 | $960.00 |
| Standard, 50% of input cached | $394.50 | $789.00 |
| Batch, no cache hits | $240.00 | $480.00 |
The cached line is 45 x $2 + 45 x $0.10 + $300 = $394.50 in November, and 45 x $4 + 45 x $0.20 + $600 = $789 in December. The batch line is 90 x $1 + 30 x $5 = $240, then double.
Two other models, same workload
Same 90 million input and 30 million output tokens, standard rates, no caching, from the same pricing page and Anthropic's Haiku 5.5 page, checked October 15, 2026. This assumes each model bills the same 1,000 output tokens per request, which will not hold exactly: tokenizers and thinking budgets differ.
| Model | Input / output per million | Nov 2026 | Dec 2026 |
|---|---|---|---|
| Gemini 4 Argon | $2 / $10, then $4 / $20 | $480.00 | $960.00 |
| Gemini 3.8 Flash | $0.75 / $3.75 through Dec 31, 2026 | $180.00 | $180.00 |
| Claude Haiku 5.5 (prompts up to 100K) | $0.10 / $0.50 | $24.00 | $24.00 |
Flash: 90 x $0.75 + 30 x $3.75 = $67.50 + $112.50 = $180. Haiku: 90 x $0.10 + 30 x $0.50 = $9 + $15 = $24. The Flash rate rises to $1.50 and $7.50 on January 1, 2027, per the same page. Price is one column; whether a cheaper model does the job is a test you have to run on your own prompts.
What moves the number
The date moves it most: the rate doubles, so everything doubles. Next come thinking tokens. Drop them from 600 to 200 per request and output falls to 18 million tokens, which is $180 in November. That is $120 less, more than caching saved in the table above.
Input size matters least here. Cutting the prompt by a third saves $60 in November on this workload.
There is also a calendar trap. The introductory rate applies to usage through November 30, so a December invoice for the same traffic is the first one at the new price. If you are building a forecast for the first quarter of 2027, use the $4 and $20 column, not the one in the launch coverage.
Batch halves the bill, but the page does not say what turnaround batch work gets, so check the Gemini docs before you move anything that has a user waiting.
Checking your own numbers
Paste a typical prompt into the calculator for a token count, with one warning. Its encodings are the OpenAI ones, and the site's rates are April 2023 ones, so use it for the input side only and expect Gemini's own count to differ. Then take the usage numbers the API returns after a day of real traffic and multiply by the table. If thinking tokens turn out to be 1,500 a request, output grows to 57 million tokens and the $480 above becomes $750 in November (90 x $2 + 57 x $10), $1,500 in December.