Gemini 4 Argon launch: price and availability
Google announced Gemini 4 Argon on 30 September 2026 and first gave it to a small group of security testers. It became generally available on the paid Gemini API on 13 October, under the model ID gemini-4-argon. It is not on the free tier.
| Per 1M tokens | Until 30 Nov 2026 | From 1 Dec 2026 |
|---|---|---|
| Input | $2.00 | $4.00 |
| Output (including thinking) | $10.00 | $20.00 |
| Cache reads | $0.10 | $0.20 |
Batch requests cost 50% less, and Priority inference costs 1.8 times the standard price. Argon supports implicit caching only: repeated prompt prefixes are billed at the cache rate automatically, but you can't create explicit caches as you can on Gemini 3.1 Pro. Google's pricing page lists one rate for Argon whatever the prompt length, where Gemini 3.1 Pro charges more above 200,000 tokens.
What the 1 December price change means
A chat or retrieval app sending 10M input and 2M output tokens a month costs $40.00 on Gemini 4 Argon at the introductory price, against $44.00 on Gemini 3.1 Pro. From 1 December the same Argon workload costs $80, about 1.8 times Gemini 3.1 Pro. The model card's worked examples above use the introductory price; from December each of them doubles. If you build on Argon now, budget for the higher price.
Long outputs are billed as output
Argon writes up to 65,536 tokens by default and up to 1,048,576 with long decode. Thinking tokens are billed at the output rate too, so long reasoning runs add up quickly: a single 200,000-token answer costs $2.00 at the introductory output price and $4.00 from December.
What Google says Argon can do
- Coding: 77.9% on DeepSWE v1.1, which Google calls a new state of the art for long-horizon software engineering.
- Knowledge work: top of the Vals Index, which covers finance, coding, legal and tax work, and first on Zapier's AutomationBench at 51.3%.
- Security: finding and patching vulnerabilities, and Google's strongest resistance yet to prompt injection.
These are Google's own claims. On Epoch AI’s independent Capabilities Index, Argon scores 167 (range 164–170), level with Claude Fable 5.1 (167) and within the margin of error of Claude Opus 5.5 and GPT-6 Astra (both 169).
Gemini 4 Argon or Gemini 3.1 Pro?
- Gemini 4 Argon for long reasoning, agents and very long outputs. Until 30 November it costs the same per token as GPT-6.1 Sol and Claude Sonnet 5.5; from December it costs twice as much.
- Gemini 3.1 Pro if you rely on explicit context caching, or want a price that isn't about to change. Google still labels it a preview model.
- Gemini 3.8 Flash for most everyday work at a fraction of either; its own promotional price ends on 31 December.
Free tier limits, context caching and grounding charges for every Gemini model are on the Gemini API pricing page.
Sources: Gemini API pricing and Gemini 4 Argon model page (checked 22 October 2026), Gemini API changelog, Google announcement, Epoch AI Capabilities Index (CC BY 4.0).
