Perplexity · OpenRouter API
Decider V1 27B by Perplexity scores 83.3 on the System One Index. Calibration error 5.6%. An earlier run, kept for reference.
| Dataset | Accuracy | Rank | Best on the board | Calibration error |
|---|---|---|---|---|
| AG News English | 92.0 | 92.5 | 6.9% | |
| SST-2 English | 96.5 | 97.0 | 1.5% | |
| SST-5 English | 60.0 | 62.5 | 4.0% | |
| DAIR Emotion English | 59.5 | 91.5 | 13.2% | |
| Banking77 English | 81.0 | 98.5 | 6.8% | |
| BoolQ English | 92.5 | 92.5 | 2.2% | |
| XNLI English English | 94.0 | 93.3 | 3.9% | |
| Typed decisions English | 73.2 | 79.0 | 9.2% | |
| XNLI French French | 90.0 | 90.7 | 4.5% | |
| XNLI Arabic Arabic | 89.3 | 89.3 | 5.0% | |
| MASSIVE scenario Arabic | 72.5 | 89.5 | 12.4% | |
| MASSIVE intent Arabic | 90.0 | 91.5 | 5.6% | |
| OSACT4 offensive Arabic | 95.5 | 95.5 | 2.9% | |
| Tweet sentiment Arabic | 72.5 | 80.0 | 6.0% | |
| MIRACL rerank Arabic | 86.5 | 90.5 | 6.0% | |
| MIRACL pairwise Arabic | 88.5 | 91.0 | 5.9% |
perplexity/pplx-decider-v1-27bRetired by Perplexity in October 2026 (OpenRouter now returns 404) and replaced by Decider V1.1 27B. Left out of the charts. Its latency is from the original run, on a different machine.
Ranks count the models charted on the leaderboard. Every model gets the same cases and the same request, under the benchmark rules.