All coding cohorts¶
Different loaded configurations; same seven-task fixture repeated three times. Latencies are descriptive observed task times, not matched general speed claims.
| Configuration | Checks | Whole solutions | Median seconds | Summed seconds | Evidence |
|---|---|---|---|---|---|
| reference | 114/123 | 15/21 | 1.427 | 27.701 | Raw / grade |
| historical-reference | 120/123 | 18/21 | 0.828 | 16.151 | Raw / grade |
| ignis | 105/123 | 15/21 | 0.585 | 11.493 | Raw / grade |
| ignis-l 4-thinking | 123/123 | 21/21 | 40.669 | 668.108 | Raw / grade |
| qat | 123/123 | 21/21 | 3.271 | 59.566 | Raw / grade |
| coding-candidate | 120/123 | 18/21 | 3.632 | 65.456 | Raw / grade |
| coding-candidate-df | 123/123 | 21/21 | 0.870 | 16.747 | Raw / grade |
| reference-matched | 120/123 | 18/21 | 0.787 | 15.375 | Raw / grade |
| reference-matched-thinking | 108/123 | 18/21 | 30.642 | 614.542 | Raw / grade |
| reference-budget 1024 | 96/123 | 15/21 | 5.146 | 160.634 | Raw / grade |
| qat-df | 120/123 | 18/21 | 0.754 | 14.882 | Raw / grade |
| qat-draft 3 | 117/123 | 18/21 | 1.140 | 23.004 | Raw / grade |
| qat-reverse | 123/123 | 21/21 | 3.286 | 59.583 | Raw / grade |
| ignis-budget 1024 | 111/123 | 15/21 | 4.702 | 99.145 | Raw / grade |
| ignis-budget 4096 | 123/123 | 21/21 | 16.639 | 305.500 | Raw / grade |
| ignis-256k-vision | 123/123 | 21/21 | 16.591 | 305.187 | Raw / grade |
| qat-text 160k | 123/123 | 21/21 | 3.268 | 59.461 | Raw / grade |
| coding-candidate-128k-recheck | 123/123 | 21/21 | 0.886 | 16.838 | Raw / grade |
| reference-restored | 114/123 | 15/21 | 1.399 | 27.544 | Raw / grade |