LLM API prices / Decisions API

OpenAI Decisions API pricing

Last checked: 2026-10-07. The official Decisions guide calls it a public beta, currently uses gpt-6-luna, and shows POST /v1/decisions. Its standard input rate is $0.10 per million tokens. The guide says only input tokens are charged: output, cache-read and cache-write token charges are zero. Regional and long-context multipliers can still apply.

Why the endpoint matters

Decisions returns typed predicate, choice or score answers. It is useful when a system must route a request among known actions or classify content against a stated criterion. It is not a general prose generation endpoint. Responses API uses the ordinary GPT-6 Luna token schedule, so its generated answer consumes billable output tokens. The two endpoints can have different costs for the same input because their outputs and intended tasks differ.

A second example: 100,000 input tokens on Decisions costs 0.1 × $0.10 = $0.01 at the base rate. The same input plus 50,000 output tokens on Responses costs $0.01 + 0.05 × $0.50 = $0.035. These are arithmetic examples, not a claim that the endpoints return equivalent content. Image input and longer prompts may need different accounting.

Choosing a billing path

Use a decision question when the desired answer is a constrained classification or selection. If the workflow needs an explanation, arbitrary text, built-in tools or a conversation, budget for Responses instead. Count actual usage from the API response and compare it with these simple calculations. Check the linked documentation again before deployment because beta features and model support can change.