Max Kan writes about AI token economics at SemiAnalysis. His coauthored analyses examine subscription limits, workload-dependent API-equivalent value and the changing capability gap between open and closed models.[1][2][3]
Start with Anthropic Subscriptions Offer 5x+ More Value Than OpenAI →
X · maxkan Contributions
Measure a plan against a workload
Kan’s subscription analysis with Andrew Megalaa and Dylan Patel examines how model choice and token type affect a plan’s apparent value. Fresh input, cached reads and generated output consume credits differently. Comparing a subscription with API prices therefore requires a defined workload, not just a total token count. The team describes testing limits across plans and monitoring changes over time.[2]
Capabilities change the economic comparison
With Evan Cloutier, Jordan Nanos and Dylan Patel, Kan also compares model capabilities by technical era, accounting for benchmark saturation.[3]
Ideas
The analysis distinguishes API-equivalent usage from the subscription payment. More cached context or another model changes the estimated value of an allowance. Its framework compares a plan, model and workload together, with changing limits included in the measurement.[2]
Tools, writing & talks
Writing, research & talks
Start here
Anthropic Subscriptions Offer 5x+ More Value Than OpenAInewsletter.semianalysis.comA workload-specific analysis of subscription limits and API-equivalent usage.
Start with the explanation of credits and token types before the headline comparison. Basic knowledge of model APIs and cached context helps interpret the charts.
Coauthor with Andrew Megalaa and Dylan Patel[2]SemiAnalysis discussion on subscription economics: a $200 Claude plan can yield about $12,000 of Opus 5.5 tokens at API pricing, versus roughly $5,000 to $6,000 on comparable ChatGPT plans, depending on workload and model.
Catalog workOriginal source- Are Open Models Catching Up?newsletter.semianalysis.com
An era-specific comparison of open and closed model capabilities.
Coauthor[3]