Model
GPT-6 Luna High
OpenAIReasoning setting: effort highClosed weights
A small, fast model of OpenAI.
- Rating
- 119895% interval 1112–1331
- Rank
- 5of 9 models
- Point share
- 70%28–12 in 40 games
- Turns / game
- 20.3
- Wrong answers
- 0.0%first answers with no legal action
- Random picks
- 0.0%after 3 wrong answers
- Output tokens / turn
- 313per saved decision with usage; reported thinking counted once
- Cost / game
- $0.01recorded token-rate accounting, not necessarily cash charged
- Seconds / turn
- 8.1 sper accepted decision
Its rating among all players
- GPT-6.1 Sol · lowest reasoning (low)1588
- Foul Play1560
- Claude Opus 5.5 · lowest thinking (low)1352
- Claude Fable 5.1 · lowest thinking (low)1339
- GPT-6 Sol · lowest reasoning (low)1330
- GPT-6 Luna · high reasoning1198
- Claude Sonnet 5.5 · lowest thinking (low)1122
- Heuristic bot1098
- GPT-6 Luna · lowest reasoning (low)1022
- Max-damage bot1000
- Claude Haiku 4.5 · high thinking967
- Claude Haiku 4.5 · no thinking816
- Random bot244
Against each opponent
Point share: the share of points won (a win is 1, a tie is ½). Expected: the score that the two ratings predict. A large gap between the two can mean that this pairing plays out differently from what one rating per player can say.
| Opponent | Rating | Games | Points | Point share | Expected |
|---|---|---|---|---|---|
| Heuristic bot | 1098 | 20 | 13 | 65% | 64% |
| Max-damage bot | 1000 | 20 | 15 | 75% | 76% |
Replays
- Against Max-damage botLost23 turns