Sage-S2.1: benchmark results
Provider: Shanghai AI Lab. Access: Open.
Aggregate Rating 1701 ± 29, rank #325 of 10489 rated models, from 96 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - MT-Bench (Japanese) - Extraction | 98 | Judge rating (1-10, x10) | 91.2 |
| Nejumi 4 - ALT - Bias | 93.44 | Score (%) | 89.4 |
| Nejumi 4 - jaster (0-shot) - JCoLA (out-of-domain) | 89 | Exact match (%) | 87.1 |
| Nejumi 4 - GLP - Function Calling | 67.96 | Score (%) | 86.1 |
| Nejumi 4 - BFCL - Live AST | 73.15 | Accuracy (%) | 83.6 |
| Nejumi 4 - GLP - Application Development | 66.41 | Score (%) | 80.8 |
| Nejumi 4 - jaster (0-shot) - ALT Japanese-to-English | 90 | COMET wmt22 (x100) | 80.2 |
| Nejumi 4 - jaster (0-shot) - JCommonsenseQA | 99 | Exact match (%) | 80.2 |
| Nejumi 4 - Toxicity - Fairness | 98.31 | Criteria met (%) | 79.9 |
| Nejumi 4 - jaster (2-shot) - JCommonsenseQA | 99 | Exact match (%) | 77.7 |
| Nejumi 4 - jaster (2-shot) - AIO | 92.9 | Character F1 (x100) | 74.2 |
| Nejumi 4 - BFCL - Multi-Turn | 52.22 | Accuracy (%) | 73.9 |
Interactive version: theaggregate.ai/model?slug=sage-s2-1 · How It Works · Data refreshed daily, snapshot 2026-11-20.