Summary
Theo of t3.gg argues that developers choosing a model for coding work can narrow a crowded field to two options: Claude Opus 5.5 and Claude Fable 5.2. The video acknowledges the wide range of current models, including GPT-6 variants, Gemini 4 Carbon, and Kimi K3.1 Pro, but contends that without specific constraints, the alternatives are hard to justify.
A large part of the discussion is about cost. Theo compares subscription value against raw API pricing, noting that a Claude subscription can deliver several thousand dollars of usage per month, and explains how the $200 plan’s limits work, including the cap on how much of the weekly allowance can be spent on Fable. By his estimate, that cap yields roughly $3,000 of Fable usage versus about $14,000 of pure Opus usage.
The video also examines latency and throughput. Opus 5.5 runs at roughly 100 to 115 tokens per second, Sonnet 5.5 around 145, and Haiku 5.5 about 255. Surprisingly, Fable 5.2 measures around 106 tokens per second in his day-to-day use, putting it on par with Opus for speed. The video also covers when each model fits best and includes a sponsor segment on Browserbase.
📺 Source: Theo – t3․gg · Published November 13, 2026
🏷️ Format: Comparison
