The short version — six things that matter
- $2.00 input, $10.00 output and $0.10 caching are introductory prices. The pricing cells read “$2.00 through November 30, 2026. $4.00 starting December 1, 2026.” On December 1, 2026 all three double.
- No free tier. Verbatim: “Free Tier: Not available”.
- Implicit caching only. Verbatim: “Implicit caching only (explicit caching is not supported)”. You cannot create or manage caches explicitly.
- The context window is a unified input+output limit of 1,048,576 tokens. Verbatim: “Context window (unified input and output token limit) 1,048,576”.
- Output pricing includes thinking tokens. Verbatim: “Output price (including thinking tokens)”. Heavy thinking raises the billed output side.
- No day-level release date is published. The only date signal is “Latest update: October 2026” on the model page, so we write “October 2026” instead of inventing a day.
Pricing (USD per 1M tokens, official pricing page)
Gemini API, Standard mode. Same values as the Google section of our pricing comparison. The right column is the forecast the official page labels “starting December 1, 2026” — it is not in effect yet.
| Item | Through Nov 30, 2026 | From Dec 1, 2026 (forecast) |
|---|---|---|
| Input price | $2.00 | $4.00 |
| Output price (thinking tokens included) | $10.00 | $20.00 |
| Context caching price | $0.10 | $0.20 |
| Free tier | Not available | Not available |
“Output price (including thinking tokens) | Not available | $10.00 through November 30, 2026. $20.00 starting December 1, 2026.”
“Context caching price | Not available | $0.10 through November 30, 2026. $0.20 starting December 1, 2026. Implicit caching only (explicit caching is not supported).” — Gemini Developer API “Pricing”, Gemini 4 Argon / Standard (checked 2026-10-21)
Note: we only carry time-limited wording that appears verbatim in the price cells. Batch / Priority rates and any long-context surcharge for gemini-4-argon were not confirmed on the official pricing page during this check, so they are marked unconfirmed rather than estimated.
Specs (official model page)
What it is positioned for (official wording)
Our reading: it targets (1) writing and fixing code, (2) hard maths and science, (3) agents that call tools repeatedly, and (4) very long generations. We found no benchmark table on the official model or pricing page, so we publish no numbers — see how to read benchmarks for why third-party harness numbers need a label.
Cost examples (our arithmetic)
Unit price × token count, calculated by us — not an official quote. Output pricing includes thinking tokens, so heavy thinking increases the output side.
| Usage | Through Nov 30, 2026 | From Dec 1, 2026 (forecast) |
|---|---|---|
| 1M input + 0.1M output | $3.00 | $6.00 |
| 100K input + 20K output | $0.40 | $0.80 |
| 1M input + 1M output (long decode) | $12.00 | $24.00 |
| 1M cache reads (implicit) | $0.10 | $0.20 |
Use the token calculator for your own token counts, and prompt caching for how cache billing works.
Caveats and what we could not confirm
- December 1 pricing is a forecast. It comes from “starting December 1, 2026” in the official price cells and does not apply before that date. We keep intro and forecast prices in separate columns.
- Explicit caching is not supported, so you cannot control cache creation or eviction.
- No free tier — a paid API key is required to try it.
- Unconfirmed: Batch / Priority rates, any long-context surcharge, official benchmark numbers, and a day-level release date. We do not invent values or dates.