Summary
Google’s Gemini 4 Argon is now available, and Sam Witteveen shares his hands-on impressions after early testing. His main question is not whether it is the best model, but why Google built it and who it is for.
He finds Argon strongest on tasks that combine visuals, code, and media, such as scripting Blender animations, generating audio, and building Three.js or shader-based scenes. In one demo, a single short prompt produced a working music maker inspired by Propellerheads’ ReBirth, with 808, 909, and 303 sounds and a step sequencer. He attributes this to heavy reinforcement learning on code, render, look, and fix loops.
On benchmarks, he notes that Argon scores 53 on the Artificial Analysis Intelligence Index, tying GPT-6 Astra and Fable 5.1 but trailing Opus 5.5 and Meta’s Muse Wonder. He does not consider it the top choice for agentic coding, saying Antigravity still lags Claude Code and Codex.
His analysis is that Google needed a credible frontier model for its enterprise customers and for consumer products like the Gemini app and NotebookLM, letting them stay inside the Google stack. Gemini 4 Argon is rolling out across the paid Gemini API, Antigravity, and the Gemini app.
📺 Source: Sam Witteveen · Published October 13, 2026
🏷️ Format: Review
