Its formula separates uncached input, cached-input reads, cache writes, visible output, and reasoning tokens. The article says output is commonly priced four to eight times above input across the current offerings it reviewed from OpenAI, Anthropic, and Google. It also notes that cache-read discounts and cache-write prices can differ by provider and cache duration.
The piece compares usage fields across OpenAI Chat Completions and Responses, Anthropic Messages, and Gemini generateContent. For OpenAI and Gemini, cached tokens are included in the reported input total; Anthropic reports the post-cache input tail separately. It says reasoning tokens are included in output for OpenAI and Anthropic, while Gemini thought tokens must be added to candidate output tokens for a comparable calculation.
The source does not provide a universal bill estimate, benchmark a particular workload, or establish that one provider is cheaper. Its stated prices and conditions were checked on 18 October 2026 and may change.