OpenAI Decisions API Pricing: Input, Cache & Credits

The Decisions API charges $0.10 per million uncached input tokens and does not charge for output tokens. It is a metered OpenAI API endpoint, not an unlimited ChatGPT feature. Cache writes cost extra relative to uncached input; cache hits cost less.

Verified November 8, 2026. Status: generally available. Supported model: gpt-6-luna. Endpoint: POST /v1/decisions.

Decisions API pricing table

USD per million tokens, Standard short-context rates. This table applies to Decisions, not every request made with Luna.

The Decisions guide sets the endpoint's input-only billing and links its cache prices to the standard Luna pricing table. Do not import Luna's ordinary $0.50/M output charge into this endpoint.

Budget examples without caching

For illustration, 100,000 requests containing 1,000 billable input tokens each total 100 million tokens: $10. At 10,000 input tokens per request, the same request count totals one billion tokens: $100. These are arithmetic examples, not measured workload costs. Include question instructions and other billed input, not just the document being classified.

There is no published per-question price in the documented token schedule. Multiple questions share one request, but their input still affects usage. Combining questions should be evaluated for task quality and actual token consumption rather than assumed to make the extra questions free.

Cache accounting without double-counting

Reusable shared input needs at least 1,024 tokens; question-specific content is outside that reusable prefix. Cache-read and cache-write counts are already included in usage.input_tokens. Separate those categories before applying rates rather than adding them to the total again.

For example, one million cache-read tokens plus one million otherwise uncached tokens costs $0.01 + $0.10 = $0.11, before any separate cache write or modifier. A one-million-token 30-minute write costs $0.125, not $0.10 plus $0.125 for the same tokens.

Free credits, trials and ChatGPT subscriptions

No unconditional Decisions-specific free quota is published in the documentation reviewed. However, eligible ChatGPT Pro and Business subscribers can activate API credits, and Decisions is explicitly covered. Linking the eligible subscription to an API organization is required. Allowances, expiry and workspace caps apply; Free, Go and Plus are not eligible for this benefit.

Credits offset metered API usage; they do not change the token rate or make usage unlimited. When credits run out, purchased credits can cover further usage. Without a remaining balance, the documented response is a credit-balance error, not an indefinite free trial.

Availability, regions and unpublished terms

Regional processing premiums and long-context input multipliers apply. OpenAI lists a 10% regional-processing uplift for eligible newer models. Confirm the applicable context tier before budgeting large requests. The endpoint guide does not publish a separate enterprise discount schedule or an endpoint-specific Batch/Flex rate; don't assume ordinary Luna processing-mode discounts apply unchanged.

ZDR and HIPAA support are conditional on customer eligibility and agreements, not automatic properties of any API account. Regional processing is documented for the US and Europe. Procurement should verify required data controls before comparing discounted rates.

When this pricing owner applies

Use Decisions for bounded classification, choices and rubric scores. Free-form answers and general model selection belong to our OpenAI API pricing owner. Other model calls, tools, hosting and application operations remain separate cost components. For implementation controls, consult OpenAI's documentation rather than treating low token cost as a guarantee of decision accuracy.

Scope: current published endpoint pricing and availability, not a model benchmark, a ChatGPT plan comparison or a claim of hands-on evaluation. Machine-readable rates are available in our endpoint pricing data.