Google DeepMind has a price on Gemini 4 Argon and almost no public checkout. The company post lists an introductory rate of $2 per million input tokens and $10 per million output tokens, with cached input at 95% off the input rate. As of October 2, that rate applies to a closed set of cyber defenders in the Fairwind Program, not to the paid API or Google AI Ultra.
Koray Kavukcuoglu, SVP at Google DeepMind, framed Argon as a model for long software work, legal and finance tasks, and cyber defense. Google said it is in the U.S. government’s voluntary pre-release access process and will widen access only after early-tester feedback on guardrails. A version without cyber guardrails is going to those defenders and to internal teams.
On the numbers Google published, Argon hit 77.9% on DeepSWE v1.1, which the company called a new high for long software-engineering tasks, and led the Vals Index across finance, coding, legal, and tax. Help Net Security, citing Google, put the model at 68% on CWE-bench v1, tied for first on that patching test, 51.3% on Zapier’s AutomationBench, and 91.7% on LVBench.
Key takeaways
- Introductory price: $2 per million input tokens, $10 per million output tokens.
- Access on October 2 is Fairwind cyber defenders, not general API or Ultra.
- Google reports 77.9% on DeepSWE v1.1 and 68% on CWE-bench v1.
Who can use Gemini 4 Argon on October 2
Not a creator with a ChatGPT or Claude subscription. Fairwind is a named program for trusted defenders. Google said developers, enterprises, and consumers come later, starting with paid API customers and AI Ultra subscribers. That sequence matters more than the benchmark line. A 68% patch score that only a security desk can call is a product constraint, not a leaderboard win a newsroom can ship on.
Thousands of Googlers are already using Argon internally for coding and research, the blog said. Outside that wall, the comparable public price point is OpenAI’s GPT-6.1 Sol, also quoted at $2 and $10 per million tokens in DevDay coverage. Argon matches that sticker and withholds the endpoint.
How labs sequence model access into a working desk is the subject of How AI is Reshaping the Content Creator Stack in 2026. Argon is not in that stack yet. The invoice rate is.
What to watch next
A date for paid API access, and whether the $2 / $10 rate survives the first public week. Until that endpoint exists, the Fairwind list is the product.