What Google announced
Gemini 4 Argon was unveiled on 30 September and has been available since 13 October 2026 through the paid Gemini API, Google AI Studio, Google's development environment Antigravity and eligible Gemini Enterprise subscriptions. The model isn't on the free tier. Google positions it for extended work on complex tasks in coding, finance, law, cybersecurity and science.
On price, Google writes in the announcement and in the API documentation: introductory price of 2 US dollars per million input tokens and 10 US dollars per million output tokens (including thinking tokens), cached input tokens at a 95 percent discount. The introductory period ends on 30 November 2026, after which the price rises to 4 and 20 US dollars. The price table gives the same date for all tiers.
What that means for a pilot project
An example with assumed figures: a process uses 100 million input and 20 million output tokens a month. At the introductory price that is 200 plus 200, so 400 US dollars. At the December price it is 400 plus 400, so 800 US dollars. The amount is small. The pattern still matters: anyone who approves the project at promotional prices and measures during the pilot gets a bill in December twice as high as planned. With larger volumes the difference becomes a real budget question.
The cost comparisons circulating in October also have an expiry date. VentureBeat puts Argon at the introductory price at half of Claude Opus 5.5 and a fifth of GPT-6 Astra. From December, per the report, Argon is level with Opus 5.5 and Meta Muse Wonder. The competitors' comparison prices come from that report; we haven't checked them with the vendors themselves.
Performance is open
Google's announcement cites first place on the Vals Index. Per VentureBeat, Artificial Analysis rates Argon at 53 points, below Claude Opus 5.5 at 58. On Terminal-Bench 4.0 Opus 5.5 leads in Google's own comparison (66.4 versus 57.4 percent). VentureBeat describes many figures as vendor-reported and not independently replicated. In addition, per the report the 1-million-token limit applies to input and output combined and needs a special API configuration. How it relates to maximum response length is open.
Our reading: a model that leads one index and trails another is not settled for your own task. What counts is how it does on your own documents and processes.
A pattern at Google
Back in early September Google released Gemini 3.8 Flash with a time-limited introductory price. According to the pricing page, the model costs 0.75 and 3.75 US dollars until 31 December 2026 and 1.50 and 7.50 US dollars from 1 January 2027. With Argon the pattern repeats with a shorter deadline. Two cases don't make a trend, but they suffice as a warning: with new Google models it pays to check the pricing documentation for deadlines before planning.
- Budget pilot projects at the price that applies after the promotion and treat the discount as a bonus.
- Before 30 November, test with a representative slice of your data whether Argon solves your task better than cheaper models. Decide on quality per euro, not list price.
- Look at the cheaper tiers: Batch and Flex cost 1 and 5 US dollars until the end of November, 2 and 10 after. For tasks without time pressure that can be enough.
- Don't hard-code model prices. Keep them in configuration and review them regularly.
- Set a cost budget limit with an alert. Introductory prices expire, usage usually grows.
- Plan for switchability so that a change of vendor doesn't first require a rebuild.
What remains open
Google doesn't explain why the price rises or whether it stays stable after 1 December. Whether other vendors adjust their prices, we don't know. We also only checked Google's own prices. The competitors' figures and the benchmarks come from one report and from vendor statements.