Architecture cost review
GPT-6 Luna Decisions Pricing: What OpenAI's Decisions API Costs
OpenAI released GPT-6 Luna Decisions on 6 October 2026. It is the same GPT-6 Luna model served as a decision model, so the only saving against plain Luna is the label you no longer pay to generate.
Direct answer
GPT-6 Luna Decisions costs $0.10 per million input tokens on OpenRouter, with free output: $50 per million 500-token decisions. That is 2.4 times Jev's $21 and ten times Decider Flash's $5. Pay it when a single state runs past 262,144 tokens, up to its 1,050,000-token window, or when your contracts and data reviews already cover OpenAI. Otherwise a cheaper decision model does the same job.
Decision summary
| Decision area | What matters |
|---|---|
| Input | $0.10 per million tokens (OpenRouter, 3 Nov 2026) |
| Output | $0, typed answers and probabilities |
| Context window | 1,050,000 tokens |
| Released | 6 October 2026 |
| Endpoint | /v1/decisions (LLM Gateway, LLM Reference) |
| Above 272K input tokens | $0.20 per million on LLM Gateway's listing; unconfirmed on OpenAI's page |
| Per 1M decisions at 500 tokens | $50 · Jev $21 · Decider Flash $5 · plain Luna with label $60 |
What you are buying
OpenRouter describes GPT-6 Luna Decisions as GPT-6 Luna served through OpenAI's Decisions API. You pass a state of text, JSON or images and typed questions, and it returns typed answers with probabilities. LLM Gateway and LLM Reference list the endpoint as /v1/decisions.
The rate is $0.10 per million input tokens, the same input rate as plain GPT-6 Luna. The difference is output. Plain Luna charges $0.50 per million output tokens, while the Decisions version charges nothing.
Against plain Luna: you save the label, nothing else
A 20-token label costs $10 per million items on plain Luna (20 million tokens × $0.50). That is the whole saving. At 300 tokens, plain Luna is $40 and Luna Decisions $30. At 500 tokens, $60 against $50. At 2,000 tokens, $210 against $200.
The bigger gain is code, not money. You stop parsing text and get a probability per option. If you already call plain Luna for labels and parse the reply, moving to the Decisions endpoint is a small change with no new vendor review.
Against Jev and the cheap decision models
Per token, Luna Decisions costs 2.4 times Jev ($0.10 ÷ $0.042 = 2.38) and 10 times Decider Flash. At 10 million 500-token decisions a month, that is $500 on Luna Decisions, $210 on Jev and $50 on Decider Flash.
We know of no independent test of Luna Decisions on its own. A 77-category intent test cited by Kai Waehner found two unnamed OpenAI models ahead of Jev, at 83.9% and 86.2% against 79.0%. It does not say either was Luna Decisions, so do not read it as one.
Very long states, and the 272K question
The 1,050,000-token window is the strongest reason to pay. A 400,000-token contract bundle is past Jev's and Decider's 262,144-token windows. On Luna Decisions it costs 400,000 × $0.10 ÷ 1,000,000 = $0.04 per decision.
LLM Gateway lists a second tier above 272,000 input tokens at $0.20 per million, double the base rate. It does not say whether that rate covers the whole request or only the tokens above 272,000. If it covers the whole request, the same 400,000-token decision costs $0.08. We could not confirm the tier on OpenAI's own page, so check your usage.cost on the first long calls.
Two cheaper models also read long states. Solar Decide (524,288 tokens, $0.05) would charge $0.02 for the same bundle. ScaleDown's Classify-1 (1,000,000 tokens, $0.04) would charge $0.016. Past 524,288 tokens, only Luna Decisions and Classify-1 are left.
When OpenAI is the cheaper choice in practice
If your legal team has approved OpenAI and nothing else, adding TypeSafe or Perplexity is a new data processor and a new review. For a low-volume decision point, say 100,000 decisions a month at 500 tokens, Luna Decisions costs $5 a month against Jev's $2.10. The $2.90 gap is smaller than an hour of anyone's time on a vendor form.
At volume the sums flip. At 50 million decisions a month the gap is $2,500 against $1,050, or $1,450 a month. That pays for the review many times over.
Monthly budgets on Luna Decisions
At a 500-token state, a month costs decisions × 500 × $0.10 ÷ 1,000,000. That is $5 for 100,000 decisions, $50 for 1 million, $500 for 10 million and $2,500 for 50 million. If you buy OpenRouter credits on the 5.5% Standard fee, 1 million becomes $52.75 and 10 million $527.50.
Long states set a different scale. A thousand 400,000-token decisions a day is 30,000 a month. At $0.04 each that is $1,200 a month, or $2,400 if the >272K tier applies to the whole request. The same work on Solar Decide, if your states fit its 524,288-token window, costs 30,000 × $0.02 = $600.
Moving from plain Luna to the Decisions endpoint
The model is the same, so the change is in the request and the reply. Your label list becomes typed questions with options, and the reply is a choice with probabilities instead of text. Drop the parser, keep the logging, and re-pick any confidence cutoff from a labelled sample, since you now get real per-option probabilities.
The saving grows with label length. A 20-token label saves $10 per million items. A 100-token answer, such as a label plus a short reason, saves 100 million × $0.50 ÷ 1,000,000 = $50 per million. You lose the reason, though, so only move calls whose text nobody reads.
Check before you commit
These checks are specific to Luna Decisions and come from the listings cited on this page.
- Read usage.cost on your first calls over 272,000 tokens to see whether the higher tier applies and to what
- Confirm your longest state fits 1,050,000 tokens with the questions added
- OpenRouter's 3 November listing has batch rows for plain Luna ($0.05 / $0.25) but none for Luna Decisions
- Price the same states on Classify-1 and Solar Decide, which cost less per token for long inputs
Key takeaways
- →GPT-6 Luna Decisions costs $0.10 per million input tokens on OpenRouter, with free output: $50 per million 500-token decisions. That is 2.4 times Jev's $21 and ten times Decider Flash's $5. Pay it when a single state runs past 262,144 tokens, up to its 1,050,000-token window, or when your contracts and data reviews already cover OpenAI. Otherwise a cheaper decision model does the same job.
- →Use Luna Decisions for states over 262,144 tokens or where only an OpenAI contract is approved; send everything shorter to Jev or Decider Flash.
- →The >272K tier comes from LLM Gateway's listing; we could not confirm it on OpenAI's own page, nor whether it applies to the whole request or only the tokens above 272,000.
What users report
Most reports say OpenAI's Decisions API costs more than Jev at $0.10 per million input tokens; a few found Luna-based options close enough on cost.
Paraphrased from public posts, with the date and a link to each original. These are other people's accounts, not our tests, and we have not verified each one.
One user pointed out the Decisions API price is $0.10 per million input tokens with free output, versus Jev's $0.042, questioning whether it's worth over 2x.
Billing · OpenAI Decisions API · Hacker News, Oct 2026
A team that tested the Decisions API against Jev at work said it cost more, had higher error rates, was slower, and gave worse answers.
Value · OpenAI Decisions API · Hacker News, Oct 2026
One user's internal benchmark found GPT-6 Luna in batch mode (20 records per batch) about 1.6x faster at 1.2x Jev's cost with no accuracy loss, though only for offline work.
Billing · GPT-6 Luna · Hacker News, Sep 2026
One user noted the Decisions API costs the same $0.10 per million tokens as prompting Luna for a classification, but is about 10x faster, so the gain is speed rather than price.
Value · OpenAI Decisions API · Hacker News, Oct 2026
A developer needing 500-600 concurrent calls said Jev's rate limits were their main constraint, so they used Clef and OpenAI's Decisions API to make up the difference.
Limits · OpenAI Decisions API · Hacker News, Oct 2026
How this page was prepared
Prices on this page come from OpenRouter's live models API, read on 3 November 2026. Cost per decision is state tokens multiplied by the input rate, because output is free on every decision model listed; small-LLM costs add the stated label length at that model's output rate. OpenRouter's credit fee is shown separately where it matters. We ran no private benchmark for these pages: accuracy and latency figures are credited to the vendor or third party that measured them, and vendor figures are labelled as vendor claims.
Frequently asked questions
How much does GPT-6 Luna Decisions cost?
$0.10 per million input tokens and $0 output on OpenRouter, checked 3 November 2026. That is $50 per million decisions at a 500-token state.
Is there a long-context surcharge?
LLM Gateway lists $0.20 per million input tokens above 272,000 tokens. We could not confirm it on OpenAI's own page, or whether it applies to the whole request.
Is GPT-6 Luna Decisions cheaper than asking GPT-6 Luna for a label?
Slightly. It saves the output tokens: $10 per million items for a 20-token label. At 500 tokens that is $50 against $60.
Why pay more than Jev?
For states longer than Jev's 262,144-token window, or when an OpenAI contract is already approved and the volume is small.
Is there a batch discount for GPT-6 Luna Decisions?
Not on OpenRouter's 3 November listing. Plain GPT-6 Luna has a batch row at $0.05 input and $0.25 output, but Luna Decisions shows only the $0.10 standard rate.
Is GPT-6 Luna Decisions a different model from GPT-6 Luna?
OpenRouter describes it as GPT-6 Luna served through OpenAI's Decisions API, so it is the same model behind a different endpoint. Both list a 1,050,000-token window and $0.10 per million input tokens.
Can GPT-6 Luna Decisions read images?
Yes. OpenRouter's listing says the state can be text, JSON or images. Among decision models that is shared with Span-02, d1-ultra, Drex Max and Clef, all of which cost less per token.