OpenAI Decisions API Pricing: What 10M Decisions Cost

Last updated: October 2026

Key takeaways

• Reported list price: $0.10 per million tokens, and $0.01 per million cached tokens.
• Reported latency: 100 to 180 ms per decision call, versus a 1 to 3 second floor before.
• A rival, TypeSafe Jev, is reported at $0.042 per million tokens, which makes OpenAI about 2.4x more expensive on list price.
• In an illustrative routing workload of 10M decisions a month, the bill ranges from about $130 (heavy caching) to $400 (no caching).
• The real saving is often latency, not tokens: faster decisions shorten every agent loop.

What is the OpenAI Decisions API?

OpenAI's post, Build more responsive apps with the Decisions API, presents it as an API for fast, focused, low-cost decisions in any application. It targets latency-sensitive, high-volume calls where a large model is overkill, such as classification, routing, and yes/no gates.

A note on sources: the OpenAI page itself was not retrievable when this was written, so the figures below come from press coverage, mainly AI Weekly's summary of the general availability launch. Confirm exact rates on the OpenAI pricing page on Calcaas or OpenAI's own site before you budget.

How much does the OpenAI Decisions API cost?

Coverage reports a headline list price of $0.10 per million tokens and $0.01 per million cached tokens. It does not split input from output, so treat the output rate as unknown. Coverage also names gpt-6-luna as the dedicated model, based on a community post, which is unconfirmed.

| Option | List price per 1M tokens | Cached per 1M tokens | Reported latency | |---|---|---|---| | OpenAI Decisions API | $0.10 | $0.01 | 100 to 180 ms per call (OpenAI claim) | | TypeSafe Jev | $0.042 | not reported | P50 about 210 ms (third-party, method unclear) |

The latency numbers were measured differently, so do not read the table as a clean head-to-head. Measure p95 on your own traffic.

Is the Decisions API cheaper than TypeSafe Jev?

No, on reported list prices. $0.10 divided by $0.042 is about 2.4x. Whether that premium is worth paying depends on what you value: OpenAI's latency claim, ecosystem fit, and reported compliance options such as HIPAA eligibility and Zero Data Retention. If your decisions are low stakes and latency is not user-visible, the cheaper option wins on cost alone.

What does 10 million decisions a month cost?

Here is an illustrative workload. Say your SaaS gates every request with a routing decision: 10 million decisions a month, each about 400 input tokens. That is 4 billion tokens a month. Output tokens are excluded because the coverage does not price them.

| Scenario | Effective rate per 1M tokens | Monthly cost (illustrative) | |---|---|---| | No caching at $0.10 | $0.100 | $400 | | 75% of tokens cached ($0.10 and $0.01 blend) | $0.0325 | $130 | | Jev at $0.042, no caching | $0.042 | $168 |

Per decision, the uncached case is $0.00004. At that level the token bill is rarely your problem. A routing layer that sends easy requests to this API and hard ones to a flagship model is where the savings live, as we show in model routing as a cost strategy.

When should you use it instead of a big model?

Use it when three things are true: the task is narrow (classify, route, approve), the call volume is high, and latency is visible to the user or compounds across steps. The reported customer numbers show why speed matters. One customer, Speak, reportedly cut time between actions from 1.4 seconds to 0.13 seconds. If an agent task makes 20 decisions, that is 28 seconds versus 2.6 seconds of pure waiting, an illustrative calculation.

Do not use it where one wrong decision is expensive and you have no fallback. Cheap decisions only stay cheap if error rates stay low, which is the lesson of cost per task versus cost per token.