AI · Oct 21, 2026
OpenAI on Google Cloud lets companies spend Google commitments on GPT-6, and that changes procurementGemini 4 Argon arrives with $2/$10 token pricing and a 1 million-token output limit, but not for the public
· Updated · Signal Desk newsroom
Analysis of reporting by Ars Technica. How we work

Ars Technica reports Google skipped a long-promised Gemini 3.5 Pro and jumped to Gemini 4 Argon, listing introductory pricing of $2 per million input tokens and $10 per million output and raising the output limit to 1 million tokens from 64,000. Access is limited to Fairwind partners such as Wiz for now.
Key takeaways
- Google skipped the expected Gemini 3.5 Pro and moved to Gemini 4 Argon.
- Introductory pricing is $2 per million input tokens and $10 per million output, with cached input discounted.
- The output limit rises to 1 million tokens from 64,000 previously.
- Internal uses include saving 300 TiB of memory via fleet telemetry and migrating C/C++ codebases.
- Google says Fairwind partner Wiz found a critical hospital-system vulnerability other models missed, without specifics.
What Ars reported
Ars Technica notes that Google never delivered the Gemini 3.5 Pro it had long promised for the summer and has instead jumped to Gemini 4 Argon, which the company says offers industry-leading performance in coding, knowledge work and cybersecurity. The model is not yet available to the public; Google's decision to start with defenders is covered in our story on Argon going first to trusted cyber defenders.
Google described internal uses, including analyzing fleet telemetry in a way that saved 300 TiB of memory and using agents to migrate C and C++ codebases. These are company-reported examples without independent verification.
Pricing and limits
Google listed introductory pricing of $2 per million input tokens and $10 per million output tokens, with a deep discount for cached input. The output limit jumps to 1 million tokens, compared with 64,000 for earlier models, which could enable much longer generated artifacts such as large code migrations or extended documents in a single response.
Because these are introductory rates for a model in limited access, procurement teams should treat them as provisional until general availability terms are published. Google has iterated rapidly on smaller models, releasing Gemini 3.8 Flash as its third Flash model in six weeks, so pricing across the lineup may continue to shift.
Early partners and safety claims
Partners in Google's Fairwind program, such as Wiz, are already using Argon. Google says Wiz used it to uncover a critical vulnerability in a hospital system that other frontier models missed, but did not share specifics, so the claim cannot be assessed. Healthcare IT teams may want vendors to describe how they are auditing systems as models with stronger cyber capabilities become available.
Ars also notes that chain-of-thought monitoring, which inspects a model's intermediate reasoning, is part of how Google says it will watch for misalignment. Public evaluations of the 1 million-token output claims, general availability pricing and any further detail on the hospital vulnerability are the next things to watch. More in our Google Gemini coverage and cybersecurity coverage.
Who feels it
- Platform engineers
- Migration and telemetry examples hint at where Argon agents may help first.
- Healthcare IT
- The reported Wiz finding is a reason to ask vendors about audits.
- Procurement teams
- Introductory rates may change at general availability.
What to watch
- More detail on the hospital-system vulnerability.
- General availability pricing versus introductory rates.
- Public tests of 1 million-token output workflows.
FAQ
How much does Argon cost?
Introductory pricing is $2 per million input tokens and $10 per million output.
What changed in output length?
The limit rises to 1 million tokens from 64,000.
Can the public use it?
Not yet; Fairwind partners have early access.
What did Wiz find?
Google says a critical hospital-system vulnerability, without specifics.
Companies: Google
More on Google
AI · Oct 14, 2026
White Circle says newer OpenAI and Google image models are easier to jailbreak than their predecessorsAI · Oct 13, 2026
After months off the frontier, Google ships Gemini 4 Argon with benchmark claims and aggressive pricingAI · Sep 22, 2026
Opus 5.5 hands risky cyber prompts to an older Claude model, and that routing is the real launch storyAI · Sep 30, 2026
The four rules in the White House AI commitment, and the missing fifth: penalties