Google DeepMind announced Gemini 4 Argon on September 30, per Google. Access starts with trusted cyber defenders through the company's Fairwind Program, then moves to paid API customers and Google AI Ultra subscribers. Google gave no date for the wider release.
Output rises to 1 million tokens per response, up from 64K. Competing frontier models cap at 128K, per MarkTechPost. Introductory pricing is $2 per million input tokens and $10 per million output, with cached input at 95% off. The standard rate doubles to $4 and $20. Google has not said when the introductory period ends.
The benchmark picture is split. Argon posts 77.9% on DeepSWE v1.1 and 91.7% on LVBench for long video, both reported as state of the art. It trails on Terminal-Bench 4.0 at 57.4% against 66.4% for Claude Opus 5.5, and on FrontierSWE v2 at 55.0% against 65.5% for GPT-6 Astra. These are launch-day figures. Independent runs come after general access.
The output ceiling changes job design more than the scores do. A 1M-token response can emit a whole migration or a full contract set in one call, where a 128K cap forces chunking and stitching. At the introductory rate, one maxed response costs $10. Google is also handing a cyber-capable model to security teams before the public, and says it is taking part in the US government's voluntary pre-release access process. We covered the last round of cuts in the Opus 5.5 and GPT-6 pricing piece.
Bottom Line
Argon's introductory price is half its standard rate, and nobody outside Fairwind can call it yet. Budget at $4 and $20. Once access opens, test long single-call outputs against your chunked pipelines, since cost and failure modes will differ most there.