~/system-one-bench / benchmarks/xnli-french
Text track · French · All benchmarks
Does one sentence follow from another, 3 options. Drex Max has the top score, 1.3 points ahead of Decider V1.2, but Decider V1.2 and 8 more are within sampling noise: a statistical tie for first.
| Rank | Model | Accuracy | 95% interval | Macro-F1 | Calibration error | Failed | |
|---|---|---|---|---|---|---|---|
| 1 | 91.3 | 86.8–95.8 | 90.7 | 3.5% | 0 | ||
| 2 | 90.0 | 85.2–94.8 | 89.3 | 5.6% | 0 | ||
| 2 | Elephant 1 BasetenTied for 1st | 90.0 | 85.2–94.8 | 89.7 | 5.4% | 0 | |
| 2 | Celeris-1 Decision CelerisTied for 1st | 90.0 | 85.2–94.8 | 89.1 | 9.1% | 0 | |
| 5 | Dexy-Decision-9B Patrick-18Tied for 1st | 89.3 | 84.4–94.3 | 88.7 | 6.2% | 0 | |
| 6 | 88.0 | 82.8–93.2 | 86.9 | 4.9% | 0 | ||
| 6 | GLiDE FastinoTied for 1st | 88.0 | 82.8–93.2 | 87.9 | 3.9% | 0 | |
| 8 | 86.0 | 80.4–91.6 | 85.4 | 4.8% | 0 | ||
| 8 | Mammoth 1 BasetenTied for 1st | 86.0 | 80.4–91.6 | 86.0 | 4.3% | 0 | |
| 10 | d1 Liquid AITied for 1st | 84.7 | 78.9–90.4 | 84.2 | 7.7% | 0 | |
| 11 | d1-ultra Liquid AI | 83.3 | 77.4–89.3 | 83.2 | 10.0% | 0 | |
| 11 | d1.1 Liquid AI | 83.3 | 77.4–89.3 | 82.9 | 7.8% | 0 | |
| 13 | Mercury Decide Inception | 82.7 | 76.6–88.7 | 82.2 | 10.5% | 0 | |
| 13 | 82.7 | 76.6–88.7 | 81.8 | 5.9% | 0 | ||
| 13 | 82.7 | 76.6–88.7 | 81.8 | 6.9% | 0 | ||
| 16 | 82.0 | 75.9–88.1 | 82.0 | 11.0% | 0 | ||
| 16 | 82.0 | 75.9–88.1 | 82.2 | 10.1% | 0 | ||
| 18 | 81.3 | 75.1–87.6 | 80.9 | 4.7% | 0 | ||
| 18 | 81.3 | 75.1–87.6 | 81.2 | 5.6% | 0 | ||
| 20 | 80.0 | 73.6–86.4 | 79.8 | 6.2% | 0 | ||
| 20 | 80.0 | 73.6–86.4 | 79.0 | 3.5% | 0 | ||
| 22 | Kev 4B Jared Palmer | 79.3 | 72.8–85.8 | 78.5 | 14.1% | 0 | |
| 23 | d1-3B Liquid AI | 78.7 | 72.1–85.2 | 78.1 | 7.0% | 0 | |
| 24 | 77.3 | 70.6–84.0 | 76.7 | 7.1% | 0 | ||
| 25 | Jev 1.13 TypeSafe | 76.7 | 69.9–83.4 | 76.7 | 12.2% | 0 | |
| 25 | Jev 1.14 TypeSafe | 76.7 | 69.9–83.4 | 76.7 | 10.7% | 0 | |
| 27 | Laya Router Convai Innovations | 76.0 | 69.2–82.8 | 75.1 | 16.7% | 0 | |
| 28 | Joey 1 Baseten | 74.7 | 67.7–81.6 | 74.7 | 11.8% | 0 | |
| 29 | Span-02 Respan | 73.3 | 66.3–80.4 | 73.9 | 11.8% | 0 | |
| 30 | Naluri-308M Jericho Siahayain-domain | 68.0 | 60.5–75.5 | 67.2 | 11.1% | 0 | |
| 31 | 67.3 | 59.8–74.8 | 66.2 | 27.8% | 0 | ||
| 31 | 67.3 | 59.8–74.8 | 65.9 | 8.1% | 0 | ||
| 33 | 66.0 | 58.4–73.6 | 66.0 | 8.4% | 0 | ||
| 33 | Kev 0.8B Jared Palmer | 66.0 | 58.4–73.6 | 63.7 | 12.7% | 0 | |
| 35 | d1-omni-600M Liquid AI | 65.3 | 57.7–72.9 | 65.2 | 15.8% | 0 | |
| 36 | 62.0 | 54.2–69.8 | 62.0 | 21.5% | 0 | ||
| 37 | 61.3 | 53.5–69.1 | 60.4 | 21.4% | 0 | ||
| 38 | Naluri-279M Jericho Siahayain-domain | 31.3 | 23.9–38.8 | 17.6 | 41.6% | 0 | |
| 39 | Classify-1 ScaleDown | 30.7 | 23.3–38.0 | 15.7 | 2.7% | 150 |
The bar is accuracy; the bracket on it is the 95% interval for 150 cases. Tied for 1st marks every model whose gap to the leader is within sampling noise (1.96 standard errors of the difference, treating the runs as independent). A failed call counts as wrong. In-domain models trained on another split of this dataset, by their maker's own account.
It counts towards the System One Index. Every model gets the same cases and the same request, under the benchmark rules.