Gemini 4 Argon costs $2 per 1M input tokens and $10 per 1M output tokens through November 30, 2026, then $4 and $20 from December 1. Output includes thinking tokens. Cached input is 95% off: $0.10 per 1M through November 30, then $0.20. Argon has been generally available on the paid tier of the Gemini API since October 13, 2026, as gemini-4-argon, and there’s no free API tier.
Below: the full rate card, a comparison with current models, four worked scenarios, cost per task, and the controls to set up before December 1. For the model itself, start with what Gemini 4 Argon is; for access, see is Gemini 4 Argon free. You can build cost assertions in Apidog and run them against Argon directly.
The price sheet
These rates come from Google’s Gemini API pricing page as of October 15, 2026. The launch post called the lower rate introductory; the pricing page now dates it.
| Gemini 4 Argon, per 1M tokens | Through Nov 30, 2026 | From Dec 1, 2026 |
|---|---|---|
| Input (Standard) | $2.00 | $4.00 |
| Output incl. thinking (Standard) | $10.00 | $20.00 |
| Cached input (implicit) | $0.10 | $0.20 |
| Batch / Flex: input, output, cached | $1.00, $5.00, $0.05 | $2.00, $10.00, $0.10 |
| Priority: input, output, cached | $3.60, $18.00, $0.18 | $7.20, $36.00, $0.36 |
What the table doesn’t show:
- No long-context tier. Argon charges the same rate at any prompt size. Gemini 3.1 Pro Preview charges $2/$12 up to 200K tokens and $4/$18 above.
- No cache storage fee. Explicit caching isn’t supported; implicit caching is on by default for prompts of 4,096+ tokens.
- Grounding. Search grounding includes 5,000 free requests a month, shared across Gemini 3+ models, then costs $14 per 1,000.
Argon vs current frontier and Gemini models
All prices are per 1M tokens. Competitor rates come from the vendors’ own pages as checked on October 1, 2026: OpenAI’s GPT-6 Astra page and Anthropic’s pricing page.
| Model | Input | Cached input | Output | Max output |
|---|---|---|---|---|
| Gemini 4 Argon (to Nov 30) | $2 | $0.10 | $10 | 65,536 default; ~1M with long decode |
| Gemini 4 Argon (from Dec 1) | $4 | $0.20 | $20 | 65,536 default; ~1M with long decode |
| GPT-6 Astra | $10 | $1 | $50 | 128,000 |
| Claude Opus 5.5 | $4 | $0.20 | $20 | 128K (300K on Batch, beta) |
| Claude Sonnet 5.5 | $2 | $0.20 | $10 | 128K (300K on Batch, beta) |
| GPT-6.1 Sol | $2 | $0.10 | $10 | 128,000 |
| Claude Fable 5.1 | $10 | $0.25 | $50 | 128K |
| gemini-3.1-pro-preview (prompt <=200K / >200K) | $2 / $4 | $0.20 / $0.40 | $12 / $18 | 65,536 |
| gemini-3.8-flash (intro / from 2027-01-01) | $0.75 / $1.50 | $0.075 / $0.15 | $3.75 / $7.50 | 65,536 |
- Argon from December 1 = Claude Opus 5.5. Same $4/$20, same $0.20 cached input.
- Argon through November 30 = Claude Sonnet 5.5 and GPT-6.1 Sol. Same $2/$10; the $0.10 cached rate matches GPT-6.1 Sol. Cursor lists Argon at the same $2/$10.
- GPT-6 Astra and Claude Fable 5.1 cost 5x Argon’s November rate and 2.5x its December rate. See Claude Fable 5.1 pricing.
- Argon’s $10 output is below gemini-3.1-pro-preview’s $12, Google’s previous top Pro model.
- Long prompts cost more elsewhere, not on Argon. OpenAI bills prompts over 272K input tokens at 2x input and 1.5x output, and Gemini 3.1 Pro charges more above 200K.
For capability alongside price, see Gemini 4 Argon vs GPT-6 Astra vs Claude Opus 5.5.
Thinking tokens bill as output
Google’s pricing page lists Argon’s output rate as including thinking, so read “$10 per 1M output tokens” as answer plus reasoning. Thinking levels are low, medium (the default) and high, and Google ran its launch evals “with the highest thinking settings,” the most token-hungry one.
Four worked scenarios
Every figure below is tokens divided by 1M, times the per-1M rate. Scenarios 1 to 3 assume no cache hits.
1. A typical API call: 20K in, 5K out
- Through November 30: 20,000 / 1M x $2 = $0.04 input, plus 5,000 / 1M x $10 = $0.05 output. Total $0.09.
- From December 1: $0.08 + $0.10 = $0.18.
- On Batch or Flex through November 30: $0.02 + $0.025 = $0.045.
At 1,000 calls a day, that’s $90 a day through November 30 or $180 from December 1. The same call costs $0.18 on Claude Opus 5.5, $0.45 on GPT-6 Astra ($0.20 + $0.25), $0.10 on gemini-3.1-pro-preview ($0.04 + $0.06), and about $0.034 on gemini-3.8-flash at its intro rate ($0.015 + $0.019).
2. A long agent turn: 200K in, 50K out
- Through November 30: 200,000 / 1M x $2 = $0.40, plus 50,000 / 1M x $10 = $0.50. Total $0.90.
- From December 1: $0.80 + $1.00 = $1.80.
GPT-6 Astra would charge $4.50 ($2.00 + $2.50). gemini-3.1-pro-preview charges $1.00 ($0.40 + $0.60), but 200K is its tier line, and longer prompts bill at $4/$18. Argon has no line: a 400K-token prompt costs $0.80 in input through November 30, twice the 200K figure and no more.
3. One maxed-out response with long decode
Because input, thinking and output share the 1,048,576-token window, a long prompt shrinks the maximum answer. With a 100K-token prompt, the most you can generate is 1,048,576 - 100,000 = 948,576 tokens.
- Through November 30: 948,576 / 1M x $10 = $9.49 output, plus $0.20 input. Total $9.69.
- From December 1: $18.97 + $0.40 = $19.37.
That call also hits rate limits: Google’s rate limits page caps Tier 1 spend at $10 per rolling 10-minute window, so a maxed call uses about the whole window through November 30 and exceeds it from December 1. For complex reasoning, Google suggests 131,072 or 262,144 output tokens; a 262,144-token answer costs about $2.62, then $5.24, before input. Background mode and timeouts are covered in Gemini 4 Argon’s 1M output tokens.
4. A 500K-token prompt sent 10 times
Argon uses implicit caching only: no cache to create, no storage fee, and 95% off cache hits on prompts of 4,096+ tokens. Say you send the same 500K-token codebase ten times with 5K tokens of output each, and the first call misses while the next nine hit the cache.
| 10 calls, 500K prompt, 5K output each | Through Nov 30 | From Dec 1 |
|---|---|---|
| Input, no cache hits (10 x 500K) | $10.00 | $20.00 |
| Input, 1 miss + 9 hits | $1.00 + 9 x $0.05 = $1.45 | $2.00 + 9 x $0.10 = $2.90 |
| Output (10 x 5K) | $0.50 | $1.00 |
| Total with cache hits | $1.95 | $3.90 |
| Total without cache hits | $10.50 | $21.00 |
Cache hits cut the input bill by 85.5%. Caching is implicit, so measure your real hit rate before you budget on nine of ten. For the Anthropic side of the same math, see Claude Opus 5.5 prompt caching cost math.
Per-token price is not per-task cost
As of October 15, Artificial Analysis lists $1.99 per task to run its Intelligence Index on Gemini 4 Argon (High) and a blended price of $1.47 per 1M tokens, with a score of 53 (#8 of 216 variants). The Decoder put the same run at $3.98 at the $4/$20 rates. In launch-week data, AA listed GPT-6 Astra (max) at the same score and $3.26 per task.
The catch is token volume. In that launch-week data, AA counted roughly 62K output tokens per task for Argon against roughly 27K for Astra at max, about 2.3x. Output per task through November 30: 62,000 / 1M x $10 = $0.62, vs 27,000 / 1M x $50 = $1.35 for Astra. At the December rates, The Decoder’s full-task figure for Argon ($3.98) lands above Astra’s $3.26, even though Argon’s per-token rates are 60% below Astra’s.
Vals AI showed the same pattern at launch: Sonnet 5.5 and GPT-6.1 Sol share Argon’s November price on paper, yet Vals listed them at $21.34 and $3.24 per test, more than 6x apart (Argon: $15.68).
Budget from your own token counts, not the rate card.
Cost controls to set up before December 1
- Cap output. Use
generation_config.max_output_tokenson the Interactions API orgenerationConfig.maxOutputTokenson generateContent. Google advises keeping the 65,536 default for standard tasks and not going below 32,768, because thinking eats the budget; lower the thinking level instead. - Choose the thinking level on purpose. The default is
medium. Test whetherlowholds quality on your task, since thinking tokens bill as output. - Send work that can wait to Batch or Flex. Both run at half the standard rate.
- Watch the spend window. Per rolling 10 minutes: $10 on Tier 1, $50 on Tier 2, $200 on Tier 3.
- Assert on usage. Every generateContent response ends with
usageMetadata. In Apidog, save the request withGEMINI_MODEL=gemini-4-argonas an environment variable, add a post-response assertion that computes cost from the token counts at the December rates, and fail the test when a call crosses your ceiling.
The arithmetic for that assertion:
cost = uncached_input / 1,000,000 x input_rate
+ cached_input / 1,000,000 x cached_rate
+ (output + thinking) / 1,000,000 x output_rate
What the price sheet leaves out
Per-model RPM, TPM and RPD limits aren’t in the docs; AI Studio shows yours. Google hasn’t published Gemini app limits for Pro versus Ultra, or confirmed Argon on Vertex AI. An invite-only Contributor tier exists at lower rates; plan on the standard ones.
FAQ
How much does Gemini 4 Argon cost per million tokens? $2 input and $10 output through November 30, 2026, then $4 and $20. Cached input is 95% off: $0.10, then $0.20.
When does the Gemini 4 Argon price change? November 30, 2026 is the last day of the lower rate. From December 1, input costs $4 and output $20 per 1M tokens.
Are thinking tokens billed as output? Yes. Google’s pricing page lists Argon’s output price as including thinking tokens.
How much does a 1M-token output response cost? About $10 through November 30 and $20 after, before input. It needs long decode, and your prompt shares the same window.
Is Argon cheaper than Claude Opus 5.5 or GPT-6 Astra? Per token, Argon from December 1 equals Opus 5.5 and sits 60% below Astra. Per task, it depends on token volume: Artificial Analysis measured Argon using about 2.3x Astra’s output tokens. For a cheaper Gemini, see Gemini 3.8 Flash pricing.
Is there a free tier for Gemini 4 Argon? No. The pricing page says Argon “is not available on the Free tier or on the Google AI Pro plan.” Pro and Ultra subscribers get it in the Gemini app, which isn’t API access.
Budget it now, measure it before December 1
Multiply your traffic’s token counts by $2/$10 and $4/$20, and you have Argon’s range. Then measure it: download Apidog, save your request with GEMINI_MODEL=gemini-4-argon, and add a cost assertion priced at the December rates. Request shapes are in the Gemini 4 Argon API guide; dates are in the release date timeline.



