Gemini 4 Argon launches globally: #1 on Vals Index, $2/$10 intro pricing, robotics variant

Gemini 4 Argon is Google's new frontier model for long-horizon reasoning across software engineering, legal, tax and scientific work. Sundar Pichai framed it as built for 'complex, long-horizon workflows'. Google says Argon helped refactor a 6-million-line Chromium codebase and raised agentic code submissions in Antigravity by more than 35%. It debuted at #1 on the Vals Index, and the Hacker News launch thread passed 1,013 points and 761 comments.

The model has unified 1M-token input and output limits. Through November 30, pricing is $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached tokens. After that it doubles to $4/$20. Alongside the main release, DeepMind published 'Gemini 4 Argon for Robotics', which extends the model to embodied control, plus a dedicated safety and responsibility post.

Argon lands one day after Meta's Muse Wonder and a day before AWS's Bedrock push. That makes this one of the densest frontier-release weeks of the year. Its price undercuts Claude Opus 5.5 and GPT-6 Astra by a wide margin. Developers cite about $0.18 per task versus $1.27 for Opus 5.5.

The skeptical read comes from a widely shared Dev.to analysis titled 'your real bill will disagree'. It argues Argon burns roughly 2x the tokens per task, which erodes the headline discount, especially after November 30. Ethan Mollick noted the model 'feels quite different' from rivals and suggested many competitors may be distilling Claude or GPT.