Hands-On Test of Gemini 4 Argon: Google's Stumbling Last-Ditch Effort

Google has made Gemini 4 Argon generally available tonight. It was first announced on October 1 and initially opened only to defensive users such as governments, critical infrastructure operators and security firms through the Fairwind cybersecurity program. After Gemini 3 Pro in November last year, Google released Gemini 3.1 Pro, and at I/O in May promised Gemini 3.5 Pro within a month. That was delayed for months because coding ability fell short of internal targets, and it was eventually cancelled. In August, Demis Hassabis stepped down as DeepMind CEO to become DeepMind chairman and Alphabet chief scientist, and Jeff Dean, who had supported Google for over two decades, departed along with Noam Shazeer. Google then relied on the Flash series. Argon is available in the Gemini app to Pro subscribers at $20 per month, but using it in Antigravity requires the $100 Ultra subscription. It offers three reasoning levels (low, medium, high), supports multimodal input and has a knowledge cutoff of March 2026. Output was originally announced as 1 million tokens but later revised to 1 million tokens combined for input and output. A single request outputs about 64K tokens, and a continuation protocol lets developers keep generating up to that limit. During the promotional period until November 30, pricing is $2 per input million and $10 per output million; afterward it rises to $4 and $20. On Artificial Analysis, Argon scores 53, tied with GPT-6 Astra, while Claude Opus 5.5 remains first. However, Argon averages about $1.99 per task, compared with $0.72 for GPT-6.1 Sol Max.