OpenAI · Closed weights · Released Sep 22, 2026
GPT-6 Luna
input $0.10/M · output $0.50/M · list openai-gpt-6-luna-2026-09-22 (Sep 22, 2026)
Aggregate score
26.8
±2.6 · #2 of 2
Accuracy
59%
89% coverage
Hallucination
72%
of missed items
Cost per task
$0.00056
$0.00092 per correct
Latency
—
Not measured at standard speed
Output tokens
1.1K
per task
Fingerprint
Aggregate and axis scores against every measured model. Ticks are the other models; the accent dot is GPT-6 Luna.
Aggregate
26.8
Knowledge
−2.7
Math
56.3
Reasoning lift
Score with reasoning on minus score with reasoning off, on the same items.
Aggregate
+57.0
Aggregate: Off −30.3, On 26.8, lift +57.0Knowledge
+15.1
Knowledge: Off −17.9, On −2.7, lift +15.1Math
+98.9
Math: Off −42.6, On 56.3, lift +98.9OffOnlift = On − Off
Sub-families
| Sub-family | Score | 95% int. | Acc. | Cov. | Halluc. | $/task | Rank |
|---|---|---|---|---|---|---|---|
| Knowledge | |||||||
| knowledge_depth | −2.7 | #2 | |||||
| Math | |||||||
| ballpark | 56.3 | #2 | |||||