Qwen2.5-Coder 32B Instruct
Zero-eval
#1BigCodeBench-Full
#1BigCodeBench-Hard
#2MBPP
+1 more
by Alibaba Cloud / Qwen Team
+
+
+
+
About
Qwen 2.5 Coder 32B was developed as a specialized coding model, designed to excel at programming tasks with 32 billion parameters specifically optimized for code. Built to understand and generate code across multiple programming languages, it serves developers requiring advanced code completion, debugging, and explanation capabilities.
+
+
+
+
Pricing Range
Input (per 1M)$0.09 -$0.89
Output (per 1M)$0.09 -$0.89
Providers4
+
+
+
+
Timeline
AnnouncedSep 19, 2024
ReleasedSep 19, 2024
+
+
+
+
Specifications
Training Tokens5.5T
+
+
+
+
License & Family
License
Apache 2.0
Base ModelQwen2.5 32B Instruct
Performance Overview
Performance metrics and category breakdown
Overall Performance
15 benchmarks
Average Score
64.9%
Best Score
92.7%
High Performers (80%+)
5Performance Metrics
Max Context Window
256.0KAvg Throughput
74.0 tok/sAvg Latency
0ms+
+
+
+
All Benchmark Results for Qwen2.5-Coder 32B Instruct
Complete list of benchmark scores with detailed information
| HumanEval | text | 0.93 | 92.7% | Self-reported | |
| GSM8k | text | 0.91 | 91.1% | Self-reported | |
| MBPP | text | 0.90 | 90.2% | Self-reported | |
| HellaSwag | text | 0.83 | 83.0% | Self-reported | |
| Winogrande | text | 0.81 | 80.8% | Self-reported | |
| MMLU-Redux | text | 0.78 | 77.5% | Self-reported | |
| MMLU | text | 0.75 | 75.1% | Self-reported | |
| ARC-C | text | 0.70 | 70.5% | Self-reported | |
| MATH | text | 0.57 | 57.2% | Self-reported | |
| TruthfulQA | text | 0.54 | 54.2% | Self-reported |
Showing 1 to 10 of 15 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+