GPT-6 Luna charges $0.10 per million input tokens and $0.50 per million output [OURS]. That is the lowest input rate of the 21 cards in this batch, and it sits under an intelligence score of 38.1, ranked 38th of 148 models by the independent evaluator — the 74th percentile of everything that evaluator measures. The gpt 6 luna api price query lands on that card. It is served by OrcaRouter.
The rate row
| Line | Rate per million | Against input |
| Input | $0.10 | 100% |
| Output | $0.50 | 500% |
| Cached read | $0.01 | 10% |
| Cache write | $0.125 | 125% |
A 5× output ratio, a 10 percent read, and the familiar 1.25× write premium [OURS]. At these absolute numbers the cache figures are almost theoretical — a million cached tokens costs a cent — but the shape matters if the workload is large.
At 500 million input and 50 million output tokens a month, the bill is $50 plus $25 — $75 [OURS arithmetic on the published rates].At the same volume, a card priced at $2.00 per input would result in an input cost of $1,000 alone.
Five rows, one of them ranked
The card publishes five benchmark rows, all from the independent evaluator [INDEPENDENT]:
| Row | Score | Rank |
| AA Intelligence | 38.1 | 38 of 148 |
| Long-Context Recall | 83.3 | — |
| SciCode | 54.6 | — |
| Humanity’s Last Exam | 38.5 | — |
| terminalbench_v4_0 | 12.6 | — |
Two of these deserve separate attention.
AA Intelligence at rank 38 of 148 is the only ranked row, and it is the one that places the model in the field. A 38.1 score independent of rank could mean anything; at rank 38 it means the model is in the top quarter of everything measured, for a tenth of a dollar per million input tokens.
Long-Context Recall at 83.3 is high. Cards in this batch range from 67 to 88.7, and 83.3 puts this budget model in the upper group — which is unusual. Cheap models usually trade context fidelity first.
The row that limits the card is terminalbench_v4_0 at 12.6, in the band occupied by the previous generation. So the model reads well and reasons at the top quarter, but does not do sustained terminal work.
What the rank means for a price comparison
The useful way to read this card is as a price-per-intelligence position rather than a capability claim.
| Card | Input / million | AA Intelligence |
| GPT-6 Luna | $0.10 | 38.1 |
| GLM 5.3 Flash | $0.15 | 41.8 |
| Gemini 3.8 Flash | $0.75 | 40.9 |
| GPT-5.6 Terra | $2.00 | 42.1 |
| Kimi K3 | $3.00 | 43.6 |
| Claude Fable 5 | $10.00 | 49.6 |
Above the cheapest tier, the intelligence index barely moves: 38.1 at ten cents, 41.8 at fifteen, 42.1 at two dollars. The card that costs twenty times more than Luna buys 4 points of intelligence.
That curve is the argument for the whole budget tier, and it is also a warning about reading a single index: the 3.7-point gap between Luna and GLM 5.3 Flash hides differences in what the models can do. GLM’s card carries 22 rows including a full multimodal set; Luna’s carries five.
Telemetry
Seven-day figures report 2,885 ms median time to first token, a p95 at the 10,000 ms ceiling, an error rate of 0.39 percent, and 192.9 output tokens per second [OURS]. Traffic was 776,826,037 tokens.
Fast generation, mid-pack start, low errors, three-quarters of a billion tokens a week. For a card at this price, that is the profile of a route doing bulk work.
The hyphens split the query set
The page earned 75 impressions and no clicks at an average position of 15.5 over 28 days [OURS]. The query list contains a small natural experiment:
| Query | Impressions | Position |
| gpt 6 luna api | 10 | 8.3 |
| gpt-6 luna api | 10 | 74.4 |
| gpt 6 luna model card | 5 | 8.4 |
| gpt 6 luna api price | 4 | 8.2 |
| gpt-6 luna | 4 | 2.0 |
| gpt-6-luna api | 4 | 5.0 |
| gpt 6 luna | 3 | 2.0 |
| gpt 6 luna price | 3 | 10.0 |
`gpt 6 luna api` ranks 8.3. `gpt-6 luna api` — same words, one hyphen inserted after “gpt” — ranks 74.4, on page eight.
Same page, same intent, 66 places apart on a punctuation difference. That is not a content problem; it is Google matching a hyphenated token against a slug that uses a different separator, and it is the kind of thing that looks like a ranking failure and is actually a naming one.
The un-hyphenated short forms rank best of all: `gpt-6 luna` at 2.0 and `gpt 6 luna` at 2.0, both on the first screen.
The takeaway
GPT-6 Luna prices at $0.10 input, $0.50 output, $0.01 cached read and $0.125 cache write per million tokens, with a 1,050,000-token context, a 128,000-token output ceiling, and AA Intelligence 38.1 at rank 38 of 148 [OURS].
It is the cheapest card in this batch and still lands in the top quarter on the one ranked measure it publishes. Its five rows say what it is not — terminalbench_v4_0 at 12.6 — and its price says what it is for. Click here