SemiAnalysis subscription test: Anthropic delivers about 5x the usable volume of GPT at the same price
Converted to API prices, Anthropic subscriptions offer roughly five times more usable volume on agentic tasks; flagship limits for Astra and Fable 5.1 are close, and the $500 plan gives only 21% more Astra than the $200 one.
CLAUDE DELIVERS 5X THE PERFORMANCE OF GPT FOR THE SAME MONEY SemiAnalysis tested how much work actually fits into AI subscriptions. They separately loaded input tokens, model outputs and cache while tracking how fast the limits drained Converted to API prices, Anthropic came out with roughly 5x more usable volume on agentic tasks Back in June the same analysts saw the opposite picture: ChatGPT Pro beat Claude Max by almost 2x The biggest gap right now is between Opus 5.5 and GPT-6.1 Sol. For the flagships, Fable 5.1 and GPT-6 Astra, the limits are almost even OpenAI's new $500 plan gives only 21% more Astra than the old $200 one On the other hand, Codex has no five-hour windows. Anthropic does, and a big allowance has to be used up between them The researchers also spotted a hidden A/B test. Usage rates can differ between Claude users, and this was never announced publicly So in terms of work volume, Anthropic's subscription is clearly the better deal right now. But how much a specific account actually gets is anyone's guess
This post quotes another post: x.com/mikadzyki_nft/status/2068619050465329483