Summary
Theo of t3.gg examines what OpenAI’s change to Codex plan usage means for developers, asking whether the $200 Codex plan is still worth it. Until recently, he says, that plan could deliver up to $12,000 a month in inference value, while Anthropic’s equivalent plan offered about $8,000. Shortly before DevDay, OpenAI’s Tibo announced a new way of calculating usage that effectively halves the dollars available. Theo’s own test yielded about $570 of usage in a week, or roughly $2,000 a month.
He explains how the subsidized plans came about, as OpenAI and Anthropic competed to make subscriptions feel generous even as models grew more capable and costly. He then compares subscription value with actual API pricing and speculates about margins. In his analysis, GPT-6.1 Sol launched at $2 per million input tokens and $10 per million output tokens, with cheaper cache reads, rather than the higher price OpenAI may have planned. He argues this squeezed margins and helps explain the new usage limits.
The video also weighs the Codex plan against Claude’s and considers whether a new $500 tier is the real motive. It includes a sponsored segment for Browserbase and its Stagehand SDK.
📺 Source: Theo – t3․gg · Published October 07, 2026
🏷️ Format: Opinion Editorial
