Summary
Google has announced Gemini 4 Argon, the first model in the Gemini 4 family. Sam Witteveen walks through what makes it notable. It is still in testing and rolling out to selected users, with broader availability expected later. Google pitches it for long-horizon coding, knowledge work and cybersecurity defense.
The headline spec is a one-million-token output limit in a single response, compared with about 64K for Gemini 3.8 Flash and 128K for Opus 5.5, Fable 5.1 and Astra. The video explains why that matters: fewer lost details when a harness has to stop and summarize, longer reasoning chains, and the ability to produce large artifacts like full module rewrites in one pass. Google’s example had Argon agents rewrite a video decoder in Rust to make it 2.7 times faster.
On independent measurement, Artificial Analysis scores Argon 53 on its Intelligence Index, about level with GPT Astra at max reasoning and one point ahead of GPT 6.1, and 23 points above Gemini 3.1 Pro Preview. The video puts weight on token efficiency, arguing that how many tokens a model needs per task acts as a hidden price. Launch pricing is $2 per million input tokens and $10 per million output tokens, a 50% launch discount, with cached input 95% cheaper.
📺 Source: Sam Witteveen · Published October 01, 2026
🏷️ Format: News Analysis
