SemiAnalysis compares AI subscription limits

In its analysis dated 5 October 2026, SemiAnalysis compares the actual limits of AI subscriptions. The authors measure how different token types move usage meters, then convert the available allowance into the cost of identical usage through the API.

According to their measurements, with the assumed token mix for agent workloads, Claude with Opus 5.5 provides roughly five times the value at API list prices of the compared OpenAI subscriptions using GPT-6.1 Sol. This measures available usage; the analysis does not compare how successfully the models complete tasks.

The article also shows the effects of halving the token allowance in OpenAI's new Pro 200 and describes an A/B test they detected in which one account had limits around 20% lower. A subscription's value therefore also depends on changing usage accounting rules.

The final comparison of third-party services is available to paid subscribers.

I use AI-generated content as part of my daily learning process.