gpt-oss 20B
Openai · open weights · released 2025-08-05 · route gpt-oss:20b · Model profile
Opening gap: +618 (95%: -445 to +1,680). The interval includes zero, so we can't yet say whether it performs differently across the variants.
Standard
- Rating
- 1,42795%: 627 to 2,226
- Record
- 1 won, 0 drawn, 0 lost of 1
- Accuracy (open positions)
- 85.9
- Accuracy (all moves)
- 90.8
- Average centipawn loss
- 54
- Blunders / mistakes / inaccuracies per 100 moves
- 8.3 / 0.0 / 4.2
- Matched Stockfish's first choice
- 42%
- In Stockfish's top three
- 62%
- Legal on the first try
- 91.7%
- Illegal / invalid
- 2 of 26 answers
- Forfeits
- 0
- Adjudicated games
- 1
- API-equivalent cost per game
- n/a
- API-equivalent cost per move
- n/a
- Seconds per move
- 969.6
- Model minutes per game
- 387.9
- Tokens per move (in / out)
- 368 / 11,555
- Reasoning tokens per move
- n/a
- Aborted on our side (not counted)
- 3
By phase
| Phase | Moves | Accuracy | Centipawn loss | Blunders /100 | Engine match |
|---|---|---|---|---|---|
| Opening | 10 | 83.1 | 69 | 20.0 | 50% |
| Middlegame | 8 | 99.5 | 40 | 0.0 | 38% |
| Endgame | 6 | n/a | 47 | 0.0 | 33% |
Chess960
- Rating
- 80995%: 110 to 1,509
- Record
- 0 won, 0 drawn, 1 lost of 1
- Accuracy (open positions)
- 78.5
- Accuracy (all moves)
- 81.9
- Average centipawn loss
- 106
- Blunders / mistakes / inaccuracies per 100 moves
- 7.7 / 7.7 / 15.4
- Matched Stockfish's first choice
- 8%
- In Stockfish's top three
- 23%
- Legal on the first try
- 76.9%
- Illegal / invalid
- 3 of 16 answers
- Forfeits
- 0
- Adjudicated games
- 0
- API-equivalent cost per game
- n/a
- API-equivalent cost per move
- n/a
- Seconds per move
- 1,536.9
- Model minutes per game
- 333.0
- Tokens per move (in / out)
- 392 / 8,507
- Reasoning tokens per move
- n/a
- Aborted on our side (not counted)
- 29
By phase
| Phase | Moves | Accuracy | Centipawn loss | Blunders /100 | Engine match |
|---|---|---|---|---|---|
| Opening | 10 | 78.5 | 106 | 10.0 | 0% |
| Middlegame | 3 | n/a | 107 | 0.0 | 33% |
| Endgame | 0 | n/a | n/a | n/a | n/a |