Qwen2.5 72B
Alibaba·
LLMsopen-weightWorkstation
Overview
Rivaled GPT-4o on key benchmarks. Set a new standard for open-weight models with strong performance across math, coding, and reasoning.
Capabilities and innovations
0.5B to 72B parameter range128K context windowStructured output supportStrong math and coding18T token training corpusImproved synthetic data pipelineBetter long-context handling
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 49% | unsourced |
| SWE-bench Verified | 23.8% | unsourced |
| MMLU | 86.1% | unsourced |
Architecture and hardware
- Parameters
- 72B total (Dense)
- Estimated VRAM at Q4
- ~42 GB, Workstation class
- Quantization formats
- GGUF, GPTQ, AWQ
- Recommended runtime
- Ollama
- License
- Apache 2.0
Links
More from Alibaba
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.