Qwen 2.5 72B Instruct Turbo: benchmark results
Together AI's FP8-quantized Turbo endpoint for Alibaba's Qwen2.5-72B-Instruct (September 2024), offered as a hosted API rather than downloadable weights. Provider: Alibaba. Released 2024-09-19. Access: API.
Aggregate Rating 1575 ± 23, rank #527 of 1656 rated models, from 32 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PARSE (Persian Zero-Shot) - Boolean Reasoning Simple | 0.94 | Accuracy (0-1; 600 yes/no reasoning questions; Persian-langu | 100 |
| PARSE (Persian Zero-Shot) - Multiple-Choice Reasoning Multi-Answer | 0.88 | Jaccard similarity of chosen and correct options (0-1; 600 f | 100 |
| PARSE (Persian Zero-Shot) - Multiple-Choice Reasoning Single-Answer | 0.87 | Accuracy (0-1; 600 four-option reasoning questions with one | 100 |
| Enkrypt AI - Safety Risk | 13.47 | Risk Score | 98 |
| Enkrypt AI - Bias Risk | 44.19 | Risk Score | 94.6 |
| PARSE (Persian Zero-Shot) - Boolean Multihop Simple | 0.95 | Accuracy (0-1; 600 yes/no multi-hop questions; Persian-langu | 90 |
| Enkrypt AI - Risk Score | 22.75 | Risk Score | 80.7 |
| PARSE (Persian Zero-Shot) - Boolean Reasoning Comparative | 0.87 | Accuracy (0-1; 600 yes/no comparative reasoning questions; P | 80 |
| PARSE (Persian Zero-Shot) - Factoid Reasoning Simple | 0.3 | Contains match (0-1; answer contains the gold string, split | 80 |
| PARSE (Persian Zero-Shot) - Multiple-Choice Multihop Multi-Answer | 0.94 | Jaccard similarity of chosen and correct options (0-1; 600 f | 80 |
| PARSE (Persian Zero-Shot) - Multiple-Choice Multihop Single-Answer | 0.88 | Accuracy (0-1; 600 four-option multi-hop questions with one | 70 |
| Enkrypt AI - Toxicity Risk | 2.18 | Risk Score | 65.8 |
Interactive version: theaggregate.ai/model?slug=qwen-2-5-72b-instruct-turbo · How It Works · Data refreshed daily, snapshot 2026-10-20.