has opened Argon to every developer on the paid tier of the API, at a launch price of $2 per million input and $10 per million output tokens through November 30, 2026. From December 1 it goes to $4 and $20, double. If you are thinking about building a chatbot or an agent on Argon, run the numbers at both prices.
Below: what was announced, whether you can use it from Spain and the rest of the EU, and two worked cost examples with every assumption spelled out.
What Google launched
Sundar Pichai announced it on X:
> “Gemini 4 Argon is here. It’s available globally today for developers and enterprises in the paid tier of the Gemini API and the Gemini platform, for Ultra subscribers in Google Antigravity, and Ultra and Pro subscribers in the Gemini app and Search.” > > Sundar Pichai (@sundarpichai) on X
The API models page calls it ’s flagship reasoning model, with a 1 million token and long outputs of up to 1 million tokens. The model ID is gemini-4-argon. One catch: according to the pricing page, Argon is not on the API free tier or the AI Pro plan. You need billing turned on to try it through the API.
If you already work with Vercel, it is also on their AI Gateway as google/gemini-4-argon, and Vercel says it passes on the provider price with no markup.
Is it available in Spain and the EU?
Yes, with one caveat. Spain and all 27 EU countries are on the official list of available regions for the API and AI Studio, and Pichai says it is available globally. That list covers the API as a whole. does not publish a separate country list per model, so we found no written confirmation specific to Argon in Spain. What the docs do say is that Argon only works on the paid tier.
What Argon costs
Standard prices per million tokens, taken from the official pricing page (last updated October 13, 2026):
| Item | Through Nov 30, 2026 | From Dec 1, 2026 | |—-|—-|—-| | Input | $2.00 | $4.00 | | Output (including thinking tokens) | $10.00 | $20.00 | | Context caching (implicit only) | $0.10 | $0.20 |
If the answer can wait, Batch mode costs half ($1 and $5 through November 30), and grounding with Search is free up to 5,000 requests a month, then $14 per 1,000.
Note that output includes thinking. A model that reasons a lot before answering spends output tokens your user never sees.
Example: a chatbot on your website
If you are not sure what a website chatbot can do today, we cover it in this article on chatbots with voice and tools. These are our assumptions, there to show the order of magnitude. They are not figures:
- 3,000 conversations a month.
- 6 customer messages per conversation.
- Each message sends the model about 4,000 tokens (instructions, history and snippets from your site) and gets back about 500 output tokens, thinking included.
That adds up to 72 million input tokens and 9 million output tokens a month. Through November 30 that is $144 for input and $90 for output, $234 a month. From December it is $288 and $180, $468 a month. We leave out implicit caching, which can trim the bill when instructions repeat, and taxes.
Example: an agent working in the background
Now an internal agent that reviews documents or drafts reports. Assumptions:
- 20 tasks a day, 22 working days: 440 tasks a month.
- Each task reads about 200,000 tokens and writes about 20,000.
That is 88 million input and 8.8 million output. At the launch price, $176 plus $88: $264 a month. From December, $352 plus $176: $528 a month. If the tasks can wait, Batch mode halves those figures.
What we would do
Measure first. A month of real traffic tells you how many tokens each conversation actually uses, and that number matters more than any assumption. If a cheaper model answers most questions just as well, keep Argon for the hard ones. And if you commission something now, ask for the quote to cover the December price, because it doubles.
If you already have a chatbot on your site and want to know what moving it to Argon would cost, or what to change so it spends less, that is what our one-off review for €150 is for: one concrete delivery, no commitment. If you would rather talk first, get in touch.