Gemini 4 Argon costs $2 per million tokens, for now, if you can get access
Key takeaways
- Gemini 4 Argon launches at $2 per million input tokens and $10 per million output, rising to $4 and $20 after an unspecified introductory period.
- On the Artificial Analysis Intelligence Index it scored 53, at $1.99 per task at the intro price and $3.98 at standard.
- Google has given no timeline for public access, and a 1 October hands-on found every model ID returned a 404.
$2 in, $10 out, per million tokens. That is Google's introductory price for Gemini 4 Argon, its new flagship model, and it matches the input and output prices Anthropic charges for Claude Sonnet 5.5. The catch is in the word introductory. Google says the standard rate is $4 and $20, double the launch price, and it has not said when the switch happens.
The other catch is that you probably cannot use it yet. Google announced Argon after months of delays and gave no timeline for a public release. A hands-on review published on 1 October tried three different model IDs and got a 404 from every one. For now, access is limited to organisations in Google's Fairwind cybersecurity programme, with paying API customers and Google AI Ultra subscribers next in line and no date given.
What Gemini 4 Argon pricing buys you
On the Artificial Analysis Intelligence Index, Argon scored 53 on high reasoning, tying for third place. That puts it level with OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 at their maximum reasoning settings, at a lower cost per task: $1.99 per index task at the introductory price, $3.98 at standard. These are independent figures from Artificial Analysis, not numbers supplied by Google.
The Batch describes Argon as a low hallucination model built for knowledge work. Set to high reasoning it had the lowest hallucination rate of the models compared, 15 percent, and it leads most finance, legal and business tests, though it still trails Anthropic's newest models on agentic coding. If that holds up, the price matters more than the index score, because knowledge work means long documents and long documents mean lots of input tokens. At $2 per million, a 100,000 token contract costs 20 cents to read. At $4 it costs 40.
Why it matters
Anthropic's Sonnet 5.5 and OpenAI's GPT-6.1 Sol also list at $2 and $10, so three frontier labs now sell a leading model at the same headline price, which looks like a market settling on a number. The real bill depends on effort settings, caching and how many tokens a model burns to finish a task, and none of that shows up on the pricing page. Our piece on Claude Haiku 5.5 at 10 cents per million tokens made the same point from the cheap end.
The introductory label is the part worth sitting with. A model priced at $2 to win customers and $4 to keep them is a pricing strategy, not a discount, and anyone building a product on Argon should budget at the higher number.
What to watch
A public release date, and whether the introductory period ends with it. Also watch whether paying API customers get access before the introductory rate ends. Until the API stops returning 404s, Gemini 4 Argon pricing is a press release number.
