What Happened
On 2026-09-30, Google announced Gemini 4 Argon, with initial access restricted to trusted cyber defenders in the Fairwind program.
Why It Matters
The new Gemini generation emphasized very long reasoning trajectories and complex professional workflows, with deployment restricted according to capability. The unusually large output budget was intended to let the model sustain a single extended reasoning trajectory. This is technically different from accepting a large input context, and the distinction is essential for a reliable historical record.
Technical Details
Google expanded the output-token limit to one million, compared with the prior 64,000. The announcement described internal coding, migration, and optimization uses. An introductory price of $2 input and $10 output per million tokens was announced; the release was not general availability.