25 MAR 2026
David asks: Asked whether Claude code subsidization changes the Kerra compute cost model
David asked an agent to research how much compute Anthropic's $200/month plan actually provides relative to listed API prices, prompted by concerns about running 100-agent research tasks.
While planning a 100-instance AWS research pipeline for enriching 30k student records, David paused to question his own cost assumptions:
can you do research to find out how much claude code is being subsidized? like, how much does a 200 dollar subscription give you in terms of compute? i heard its something like 5k usd of compute? do research on latest reporting (high quality) and give me the low down. do the same for codex and openai
The question arose because the agent had estimated costs in API-tier dollars, but David expected the subscription path to be on a completely different cost curve. The answer shaped how he planned to scale the student-research pipeline — using subscription compute rather than raw API billing.