Gemini 4 Argon is priced to undercut Claude—but still not callable
Google's new flagship posts 77.9% on DeepSWE with 1M-token outputs and $2/$10 pricing, yet it's gated to cyber defenders—paid API and AI Ultra are 'next.'
Copy markdown
$2/$10 pricing lands under Claude Opus 5.5
Argon launches at $2 per million input tokens and $10 per million output, with cached input discounted ~95%—below Claude Opus 5.5's $4/$20. That's an introductory rate (list price later rises to $4/$20). Max output also jumps to 1M tokens per request from 64K, so whole-repo refactors and long agent runs stop truncating mid-answer.
The catch: cyber defenders first, no public API date
Access rolls out through Google's Fairwind program for 'trusted cyber defenders'—reportedly without cyber guardrails—while paid API customers and Google AI Ultra subscribers are only 'next,' with no date given. Until then the callable Gemini is 3.8 Flash (~$0.30/$1.50). Budget Argon into roadmaps, but don't wire it into production yet.
The benchmark sheet, if the API opens at that price
Google reports 77.9% on DeepSWE v1.1 (real-world software engineering), 68% on CWE-bench v1 (tied #1 for fixing vulnerabilities), 91.7% on LVBench (long video), and a 51.3% AutomationBench lead, plus reported wins over GPT-6 Astra on several coding and knowledge tasks. At $2/$10, that would make it a serious default for coding agents.