OpenAI Now Prices a Lead-Routing Decision at About $0.0001. What Does Your Vendor Charge per AI Action?

One hundredth of a cent. That is what OpenAI now charges for the model to route a 1,000-token lead.

OpenAI opened the Decisions API in public beta on Oct 6 2026 (OpenAI changelog). You send a record and a question. It sends back a probability, a pick from your list, or a score. It runs only on gpt-6-luna and bills input tokens only: $0.10 per million, with no output, cache-read, or cache-write charges (Decisions guide).

Most of the AI in a revenue stack does exactly this kind of job. Route the lead. Score the fit. Tag the reply. Flag the risk. Many vendors bill those steps per action or per credit. You now have a public floor for the model cost under them.

What OpenAI shipped

• Endpoint: POST /v1/decisions, public beta. OpenAI expects general availability "in the coming weeks."
• Three answer types: predicate (probability a condition is true), choice (one option from a fixed set, with probabilities and confidence), score (position on ordered levels).
• Price: $0.10 per million input tokens on gpt-6-luna. Output, cache reads, and cache writes cost nothing.
• Still billed extra: regional processing premiums and long-context input multipliers.
• Speed: OpenAI says about 10x faster than the Responses API. That is a vendor claim with no published test setup.
• Data: Zero Data Retention and HIPAA for eligible customers. Processing in the US and Europe.
• Inputs: text and images. Images must be inline base64; hosted image URLs and file IDs are not accepted.

The math a buyer should run

A lead record with enrichment fields and the question runs about 1,000 input tokens. At $0.10 per million, that is $0.0001 per decision. A million decisions a month is about $100 in model cost. The same model on the Responses API also bills output at $0.50 per million (OpenAI pricing), so a 50-token answer adds about $25 per million calls.

Now the vendor side. Salesforce lists Flex Credits at $500 per 100,000 and prices one Agentforce action at 20 credits, or $0.10 (Agentforce pricing). A million standard actions is $100,000 at list.

An Agentforce action does more than a bare classification. It runs inside your CRM with data access, permissions, guardrails, and an audit trail. That work is worth something. The point is narrower: when the AI step is a yes or no, a pick-one, or a score, the model is now a rounding error. Whatever you pay above about $0.0001 per decision is platform, margin, and risk. Know how much of each you are buying.

The same logic applies to any tool that sells AI on credits: enrichment, sales engagement, support, intent. It is the question behind who audits the agent meter and the Flex Credits vs conversation breakeven.

Where this shows up in a revenue stack

• Inbound lead routing: which segment, region, or owner gets the lead.
• ICP fit score: score the account against your tiers before a rep touches it.
• Reply triage: positive, objection, out of office, unsubscribe, wrong person.
• Call and meeting disposition: next step booked, yes or no; competitor named, yes or no.
• Renewal risk flag: score account notes and tickets against a churn rubric.
• Outbound compliance check: does this email make a claim legal has banned.

Order Form lines for vendor AI actions

1. Split the AI line. For each AI step, model cost and platform fee as separate numbers, per action or per credit.
2. Per-decision rate. Steps that only route, score, or tag get their own rate, below the general action rate.
3. Bring your own key. At volume, the vendor runs classification on your OpenAI key and you pay OpenAI direct. Put the volume threshold in writing.
4. Beta reopener. The Decisions API price is a beta price. If a vendor builds on it, tie your rate to the GA price and get a reopener if GA lands lower.
5. Input size cap. Long-context multipliers apply. Cap tokens per record and name who pays when a record runs long.
6. Region line. Regional processing carries a premium. Name the region and the premium on the order form.
7. Accuracy test. A labeled set of your own records, the thresholds, and who pays to fix misroutes.
8. Spend cap. A project hard limit on any key that runs decisions. Same controls as the Grow tier checklist. Partner-app allowance burn has its own lines in the SIWC and Marketplace Order Form.

Who owns which line

• Procurement: the split AI line, per-decision rate, bring-your-own-key threshold, and beta reopener.
• RevOps and GTM engineering: input size, region, the labeled test set, and thresholds.
• Revenue Finance: the spend cap and chargeback by key.

Soft opens

• This is a beta price. OpenAI has not said whether GA pricing will match it.
• The 1,000-token record and 50-token answer are our assumptions. Your records may run shorter or longer.
• The 10x speed claim comes from OpenAI with no published test setup. Measure your own latency.
• Agentforce list prices exclude discounts, pre-commit terms, and per-user editions with pooled credits. Your contract rate may differ.
• A rival decision model, TypeSafe's Jev, lists $0.042 per million input tokens, text only, per CellCog. Neither vendor has published head-to-head accuracy.

Sources

1. OpenAI, API changelog, Oct 6 2026 entry: developers.openai.com/api/docs/changelog
2. OpenAI, Decisions guide, pricing and availability (retrieved Oct 7 2026): developers.openai.com/api/docs/guides/decisions
3. OpenAI, API pricing (retrieved Oct 7 2026): developers.openai.com/api/docs/pricing
4. OpenAI Developer Community, Decisions API public beta announcement (Oct 6 2026): community.openai.com
5. Salesforce, Agentforce pricing and usage example (retrieved Oct 7 2026): salesforce.com/agentforce/pricing
6. CellCog, OpenAI Decisions API: Price, Question Types, vs Jev (Oct 6 2026): cellcog.ai

New Order Form decoders land in the weekly issue first. Subscribe.