Gemini 4 Argon

Gemini 4 Argon is Google DeepMind's new frontier model, unveiled 2026-09-30 and billed by Google as its most powerful model yet — the first new frontier Gemini since Gemini 3.1 Pro. Text, image, and video input with text output; a 1,048,576-token (1M) context window and a greatly expanded output budget (Google cites up to roughly 1,000,000 output tokens, up from the prior ~64K). It is a reasoning model with extended thinking. Proprietary, weights not published. Rolled out in stages — first to "trusted cyber defenders" through Google's Fairwind Program, then to paid API customers and Google AI Ultra subscribers. Introductory API pricing is $2 / 1M input ($0.10 cached input, a ~95% discount) and $10 / 1M output, reverting to $4 / $20 once the introductory period ends. Google-reported benchmarks top several leaderboards: DeepSWE 1.1 77.9% (#1), FrontierSWE V2 55% (#1), LAB-Bench 2 88.8% (#1), LVBench video understanding 91.7% (#1), and Gray Swan prompt-injection robustness 0.7% (best); Artificial Analysis measures an Intelligence Index of 53, well above the ~26 median for reasoning models. Parameter count, architecture, and knowledge cutoff are undisclosed. Figures are vendor/third-party-reported and had not been independently re-run at launch.