Gemini 4 Argon

Gemini 4 Argon is the first Gemini 4 model to reach developers, and the gap between announcement and access is part of its story. Google unveiled it on September 30, 2026 and shipped it only to trusted cyber defenders in its Fairwind Program, publishing a price before anyone outside that group could buy it. The Gemini API changelog records general availability on October 13, 2026 as gemini-4-argon, and the price list now carries the rate Google promised: an introductory $2 per million input tokens and $10 output through November 30, 2026, rising to $4 and $20 on December 1, with context caching at $0.10 and then $0.20, implicit caching only, and no Free tier. The headline spec needs reading twice. The 1,048,576 token window is unified: input, thinking, and output all draw on the same budget, so a long prompt and a long answer compete rather than add. Default synchronous output is 65,536 tokens, and reaching the full million requires long decode, which is the feature Google led with and is a ceiling rather than a default. Input spans text, image, video, audio, and PDF, and output is text only. Google positions Argon at complex coding, agentic work, math, and science, and in its announcement text reported 77.9 percent on DeepSWE v1.1, 51.3 percent on AutomationBench, 91.7 percent on LVBench, and 68 percent on CWE-bench v1. Independent pre-release measurement is unusually good for a model this new: as of September 30, 2026 Vals placed Argon first on the Vals Index at 68.90 percent, ahead of Claude Sonnet 5.5 at 67.04 and Claude Opus 5.5 at 66.97, and measured 57.58 percent on Terminal-Bench 4.0, fifth on that board; Artificial Analysis scored it 53, level with GPT-6 Astra. Those runs predate GA, so treat them as a floor on a model that has since been served more widely. At $2 and $10 it lands on the same list price as GPT-6.1 Sol and Claude Sonnet 5.5 while scoring above both on the Vals Index, which makes it the Gemini model to evaluate rather than the Flash tier Google spent the autumn iterating.