gpt-oss 20B
Openai · open weights · released 2025-08-05 · route gpt-oss:20b · Model profile · Chess results
Results
- Score
- 4095%: 16 to 64 · 5 of 42 grids so far
- Grids solved
- 2 of 5
- Blank cells filled correctly
- 74.7%
- Wrong cells / empty cells / changed a given / no grid
- 0 / 0 / 2 / 1
- Broke the rules (complete grid repeating a digit)
- 0
- Near misses (three cells or fewer wrong)
- 0
- Average stated confidence
- 100%
- Average confidence when it had not solved it
- 100%
- Sure but wrong (90% or more, unsolved)
- 2
- Calibration (Brier, lower is better)
- 0.500
- API-equivalent cost per grid
- n/a
- API-equivalent cost per solved grid
- n/a
- Seconds per grid
- 2,702
- Tokens per grid (in / out)
- 412 / 24,740
- Reasoning tokens per grid
- n/a
- Retries for an unreadable reply
- 1
- Aborted on our side (not counted)
- 98
By tier
| Tier | Needs | Solved | Output tokens | Seconds |
|---|---|---|---|---|
| Cozy | Singles only | 0 of 1 | 8,212 | 753 |
| Gentle | Scanning | 1 of 1 | 18,046 | 653 |
| Steady | Locked candidates | 1 of 1 | 16,661 | 3,346 |
| Tricky | Hidden pairs and triples | 0 of 1 | 64,841 | 5,313 |
| Devious | Wings and fish | 0 of 1 | 15,941 | 3,445 |
| Expert | Deep wings and fish | 0 of 0 | n/a | n/a |
| Beyond | No human technique finishes it | 0 of 0 | n/a | n/a |
Cells by the technique they needed
Each blank cell is tagged with the simplest technique that fills it, given everything a solver could fill before it.
| Technique | Cells | Filled correctly |
|---|---|---|
| Last free cell | 60 | 93.3% |
| Scanning | 97 | 96.9% |
| Locked candidates | 37 | 100.0% |
| Wings and fish | 20 | 70.0% |
Every answer
Cozy #1Changed a given digit 46 of 50 cells100% sure · 8,212 tokens · localGentle #1Solved 56 of 56 cells100% sure · 18,046 tokens · localSteady #1Solved 53 of 53 cells100% sure · 16,661 tokens · localTricky #1No readable grid 0 of 55 cells64,841 tokens · localDevious #1Changed a given digit 46 of 55 cells100% sure · 15,941 tokens · local