Developer review of Argon (Gemini 4): strong but expensive, still behind on coding agents
A developer says Argon is a big improvement over the Gemini 3 family but very costly to serve and still behind Anthropic and OpenAI on long-run coding agent tasks, so developers won't switch.
My read on the Argon rollout impressions: 1. A big improvement over the Gemini 3 family BUT 2. Clearly an EXTREMELY high cost/expensive model to serve, judging by the AI plan changes and user usage examples. This is a problem for model diffusion and attempts to gain share 3. Still well behind in the most important area: long run coding agent tasks, so developers won't switch. Benchmarks don’t tell story about extent of the gap. Developer impressions do. 4. Will fall much further behind as Anthropic and OpenAI release new models very soon and RSI virtuous cycle takes off for those two leaders.