Qwen2.5 72B Instruct
Zero-eval
#1MT-Bench
#1AlignBench
#3MBPP
by Alibaba Cloud / Qwen Team
+
+
+
+
About
Qwen 2.5 72B was developed as the flagship text model in the Qwen 2.5 series, designed to provide advanced language capabilities with 72 billion parameters. Built to compete with frontier models in reasoning, coding, and general language tasks, it represents Qwen's most capable instruction-following model in this generation.
+
+
+
+
Pricing Range
Input (per 1M)$0.35 -$1.20
Output (per 1M)$0.40 -$1.20
Providers4
+
+
+
+
Timeline
AnnouncedSep 19, 2024
ReleasedSep 19, 2024
+
+
+
+
Specifications
Training Tokens18.0T
+
+
+
+
License & Family
License
Qwen
Performance Overview
Performance metrics and category breakdown
Overall Performance
14 benchmarks
Average Score
77.4%
Best Score
95.8%
High Performers (80%+)
9Performance Metrics
Max Context Window
139.3KAvg Throughput
54.0 tok/sAvg Latency
0ms+
+
+
+
All Benchmark Results for Qwen2.5 72B Instruct
Complete list of benchmark scores with detailed information
| GSM8k | text | 0.96 | 95.8% | Self-reported | |
| MT-Bench | text | 0.94 | 93.5% | Self-reported | |
| MBPP | text | 0.88 | 88.2% | Self-reported | |
| MMLU-Redux | text | 0.87 | 86.8% | Self-reported | |
| HumanEval | text | 0.87 | 86.6% | Self-reported | |
| IFEval | text | 0.84 | 84.1% | Self-reported | |
| MATH | text | 0.83 | 83.1% | Self-reported | |
| AlignBench | text | 0.82 | 81.6% | Self-reported | |
| Arena Hard | text | 0.81 | 81.2% | Self-reported | |
| MultiPL-E | text | 0.75 | 75.1% | Self-reported |
Showing 1 to 10 of 14 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+