holy sh*t

holy sh*t

More

Summary

Google DeepMind’s new Gemini 4 Neon, a medium-sized model, has surprised Theo of t3.gg, who calls it possibly his favorite model right now and the first real competitor to Opus 5.5. After pushing it hard for days, he says the rough edges are far fewer than he expected.

The video starts with Google’s announcement, which places Neon third overall with a score of 61 on the Artificial Analysis Intelligence Index. Theo points to benchmark results that put it above Opus 5.5 at a lower price and close to Fable 5.2, with the gap narrowing further in code mode. He also notes that Google subsidizes generous usage in its Gemini subscription plans.

The core of the review is real agentic work in the T3 Code repository. Theo describes Neon handling a long task that earlier Gemini models would have fumbled, then diagnosing a rendering bug in the Anti-Gravity adapter, working around it with screenshots, and filing a pull request through MCP tools. The one problem he reports is that it failed to trigger his custom skills when asked. He also discusses why he thinks the release turned out so well, and what it means for his view of Google compared with OpenAI and Anthropic.


📺 Source: Theo – t3․gg · Published November 20, 2026
🏷️ Format: Review

1 Item

Channels

1 Item

Companies