Gemini 4 Argon reaches developers: $2/$10 API, million-token outputs
Google's frontier coding model opens up at state-of-the-art SWE scores — as Codex CLI users find a self-updating daemon running after every session.
Copy markdown
A frontier coding model you can call today
Argon tops DeepSWE v1.1 at 77.9% (a new state of the art) and ties for first on CWE-bench v1 at 68%, with the output cap lifted to 1M tokens, up from 64K. It's rolling out to paid API customers and Google AI Ultra first, with wider access to developers and consumers reported to follow.
The pricing: $2/$10 now, $4/$20 when the intro ends
Introductory API rates are $2 per million input tokens and $10 per million output, with cached input discounted 95% — then they double to $4/$20 once the promo window closes. Cheap enough to trial long-context coding agents now, but model the step-up before you wire it into production.
Heads-up: Codex CLI 0.157 runs a daemon after every session
Since v0.157.0, starting an interactive `codex` session auto-starts a managed background daemon that keeps running after you close the window, copies a ~374MB package into ~/.codex/, and checks for updates hourly. Users are pushing back; turn it off with `codex features disable daemon_auto_start` (or set daemon_auto_start = false in config.toml).