Reasoning Token Billing

OpenAI o1/o3/o4 emit extensive chain-of-thought before the final answer · those "reasoning tokens" are hidden from the user but billed at the output rate. A reasoning query that returns a 200-token answer might consume 3,000 reasoning tokens, making the effective cost 16× the visible answer. Anthropic Claude's extended thinking mode works similarly. DeepSeek R1 made thinking tokens visible · you can audit them.

Reasoning token pricing forces app builders to cap reasoning budgets. OpenAI added a `max_output_tokens` parameter that hard-caps total tokens; most APIs added `reasoning_effort` (low/medium/high) to tune budget. Typical reasoning-to-answer ratios: math-heavy tasks 20:1, simple Q&A 2:1. For high-traffic apps, rolling reasoning into an initial cheap pass and falling back to a reasoning model on hard cases is now standard practice.

Reasoning token opacity is controversial · OpenAI initially hid both the count and the content; community pressure forced release of token counts (not content) for o1 in early 2025. Transparency pricing: DeepSeek R1 ships visible reasoning; Anthropic shows budgets + counts; OpenAI still hides content. Pricing traps: some providers (non-OpenAI) charge reasoning tokens at input rate, making them effectively free; OpenAI and Anthropic charge at output rate. Always check the doc.