Theo Browne challenges an estimate based on multiplying an aggregate token count by one assumed price. He explains that input, output, cache-write and cache-read tokens can have different prices, and says the public total does not reveal their mix. His estimate of another developer’s costs is explicitly speculative.
Theo Browne compares benchmark prices at similar scores over time and argues that equivalent measured capability has become cheaper. He acknowledges that benchmarks do not capture every practical outcome, while arguing that newer agent workflows can shift work from the developer to the model.
Theo Browne describes his own small benchmark comparison and distinguishes experimental heavy usage from a typical user budget. His forward-looking projections about future prices and productivity are predictions rather than verified results.
Watch on YouTube
