Gemini 4 Argon: frontier model, limited early access
Google announced Gemini 4 Argon for long-horizon software engineering, legal/finance knowledge work, and cybersecurity defence. Introductory API pricing is $2 per million input tokens and $10 per million output (cached input 95% off); after the intro period that rises to $4 / $20. Output context expands to 1M tokens (from 64K). Google cites internal wins (e.g. C++→Rust migrations, memory optimisations) and strong scores on DeepSWE, Vals, and cyber benches — vendor numbers, not independent audits. Broad access is not open yet: Argon is rolling first to trusted cyber defenders via the Fairwind Program while Google runs U.S. voluntary pre-release review; paid API and AI Ultra come “as soon as possible.”
HN discussion (~900 points, ~620 comments): one camp is excited by leapfrogging and recent Gemini coding/sysadmin anecdotes; another says “announced, not shipped” and criticises Google’s delayed consumer rollout and Antigravity/agy harness friction; a third debates whether anyone still has a moat. Soft consensus: impressive claims, but most builders cannot use Argon today.
Sorted take: treat Argon as a watch-list model for coding and cyber defence — note the intro price and Fairwind gate, and only plan migrations once API access is real for your account.
Model page:Gemini 4 Argon (batch) →
Why it matters: If you budget model spend for coding agents or security tooling, lock the $2/$10 intro vs $4/$20 post-intro figures into your forecast and wait for API/Ultra access before swapping production workloads.
Google ↗Discussion on Hacker News ↗AI-assisted, reviewed by Sorted AI
▲ 1
0 comments