Overview
- The model, which Google announced on Wednesday, Sept. 30, is entering a phased release that starts with the Fairwind Program and participation in the U.S. government’s voluntary pre-release access process.
- Google published benchmark tables showing Argon tops or ties competitors on most tests, but some internal Googlers say the model’s high scores do not always match messy, real‑world coding work.
- Argon raises output capacity to an industry‑leading 1 million tokens to sustain long, multi‑step reasoning and launches with introductory API pricing of $2 per million input tokens and $10 per million output tokens before higher post‑intro rates.
- Google trained Argon for cyber defense so it can autonomously find, validate and patch vulnerabilities, and security partner Wiz used the model in Scan for Good to identify a critical flaw in healthcare software.
- Thousands of Googlers are already using Argon for internal tasks that Google says freed hundreds of terabytes of data‑center memory and sped up quantum work, and the staged, gated release follows an industry trend of tighter monitoring for frontier models.