Comparison

GPT 5.6 Luna at 0.3×: when the higher multiplier still costs less

Luna is billed at 0.3× while its two siblings are 0.1×, and it is still the cheapest per-token id in the GPT 5.6 family. The multiplier is not the price — here is the arithmetic.

The multiplier is a discount, not a price

A rate multiplier is applied to a model’s own official per-token rate; it is not a price by itself. Reading 0.3× next to 0.1× and concluding Luna is the expensive one inverts the actual bill, because the three GPT 5.6 ids start from very different list rates.

Sol lists at $5 per million input and $30 output, Terra at $2 and $12, Luna at $0.20 and $1.20. Apply each multiplier and you get $0.50 / $3.00 for Sol, $0.20 / $1.20 for Terra, and $0.06 / $0.36 for Luna.

So Terra costs 3.33× Luna per token and Sol costs 8.33× — despite both carrying the smaller multiplier. One way to keep it straight: Luna’s official list rate happens to match Terra’s post-discount rate, and Luna then discounts from there.

All three GPT 5.6 ids take the same request shape on /v1/responses.

curl https://token-share.app/v1/responses \
  -H "Authorization: Bearer $TOKEN_SHARE_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.6-luna",
    "input": "Classify this log line as retryable or fatal: connection reset by peer"
  }'

What Luna does not give up

The cheaper id in a family is usually the one with a shorter window or a lower output ceiling. Luna is not: it carries the same 372,000-token context and the same 128,000-token output ceiling as Sol and Terra, supports text, tool calling and thinking, and was released the same day, 2026-07-09.

It answers on the same /v1/responses route with the same Authorization: Bearer header, so moving traffic to it does not touch your request-building code. Everything the catalog records about the three ids matches except the per-token rate.

That leaves output quality as the only axis the catalog is silent on — which is the axis you have to measure yourself.

Where the saving is real and where it is noise

At $0.06 / $0.36 per million, one session sending 100,000 tokens in and 5,000 out costs about $0.008 on Luna against $0.026 on Terra. The absolute difference is under two cents. On a few hundred calls a month, choosing on price here is optimizing noise.

At a million such calls the same gap is roughly $18,000, and the picture inverts: the per-token rate becomes the dominant term in the bill and everything else is rounding. Batch classification, log triage, tight tool loops and anything you run per event live in that regime.

The honest test is retry rate rather than rate card. Luna is 3.33× cheaper than Terra per token, so it stays cheaper as long as it does not need more than three attempts where Terra needs one. Sample a few hundred real requests through both, count how many outputs you had to redo, and compare total spend rather than unit price. If the retries land under that ratio the saving is real; if they do not, the cheaper id is buying you a worse bill.

Common questions

Does a 0.3× multiplier mean Luna is discounted less than Sol and Terra?

Yes — proportionally, Luna keeps more of its official rate than the 0.1× ids do. That is a statement about the discount, not about the bill. Because Luna’s official rate is $0.20 / $1.20 per million against Terra’s $2 / $12, the post-discount price still comes out 3.33× lower.

Is Luna the cheapest OpenAI id in the catalog?

Among the OpenAI text ids, yes on per-token rate: Luna works out to $0.06 / $0.36 per million, below gpt-5.4-mini at $0.075 / $0.45 — even though gpt-5.4-mini carries the lower 0.1× multiplier. Luna also has the larger context of the two, 372,000 tokens against 272,000.

Does the cheaper rate come with a smaller context window?

No. Luna carries the same 372,000-token context and 128,000-token output ceiling as gpt-5.6-sol and gpt-5.6-terra, with the same text, tool-calling and thinking capabilities and the same /v1/responses route.