Model comparison
Qwen2.5-Coder-32B vs Qwen2-72B
Qwen2.5-Coder-32B is the stronger model overall, scoring 33.4 to 30.0 on the Noometry Index.
Last verified . 16 shared benchmarks.
Alibaba (Qwen)Summary
- They share 16 benchmarks with published results for both. Qwen2.5-Coder-32B scores higher in 5 categories and Qwen2-72B in 3 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where Qwen2.5-Coder-32B leads 33.4 to 21.2.
- The biggest single-benchmark swing is BigCodeBench Instruct: 49% for Qwen2.5-Coder-32B and 38.5% for Qwen2-72B.
Side by side
| Qwen2.5-Coder-32B | Qwen2-72B | |
|---|---|---|
| Provider | Alibaba (Qwen) | Alibaba (Qwen) |
| Noometry Index | 33.4 | 30.0 |
| Released | 2024-09-18 | 2024-06-07 |
| Weights | Open | Open |
| Context window | 33K | — |
| Max output | 29K | — |
| Input $ / M tokens | $0.66 | — |
| Output $ / M tokens | $1 | — |
| Results tracked | 31 | 26 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Qwen2-72B leads
Qwen2.5-Coder-32B: 22.6 (#333), Qwen2-72B: 29.1 (#310)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| BigCodeBench Instruct | 49% | 38.5% |
| LMArena Coding | 1276 | 1196 |
| BigCodeBench Complete | 58% | 54% |
| SWE-bench Verified (bash only) | 9% | — |
| Aider Polyglot | 16.4% | — |
| WeirdML | — | 11.3% |
| LiveBench Coding | 56.9% | — |
| HumanEval+ | 87.2% | — |
| MBPP+ | 77% | — |
Agentic & Tool Use Not comparable
Qwen2.5-Coder-32B: —, Qwen2-72B: 17.0 (#146)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| TheAgentCompany | — | 1.1% |
| METR Time Horizons | — | 29.9% |
Reasoning Qwen2-72B leads
Qwen2.5-Coder-32B: 21.2 (#225), Qwen2-72B: 23.2 (#181)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| LMArena Hard Prompts | 1251 | 1191 |
| Epoch Capabilities Index | 119.49 | 125.28 |
| LiveBench Reasoning | 42.1% | — |
| LiveBench Data Analysis | 49.9% | — |
| HellaSwag | 83% | — |
| LiveBench | 46.2% | — |
| WinoGrande | 80.8% | — |
Math Qwen2.5-Coder-32B leads
Qwen2.5-Coder-32B: 33.3 (#204), Qwen2-72B: 30.2 (#236)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| LMArena Math | 1251 | 1235 |
| LiveBench Math | 46.6% | — |
| MATH Level 5 | — | 39.1% |
| GSM8K | 93% | — |
Knowledge Qwen2.5-Coder-32B leads
Qwen2.5-Coder-32B: 33.4 (#203), Qwen2-72B: 21.2 (#275)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| LMArena Expert | 1221 | 1171 |
| MMLU | 79.1% | 82.4% |
| GPQA Diamond | — | 40.8% |
| ARC (AI2) Challenge | 70.5% | — |
Multilingual Qwen2.5-Coder-32B leads
Qwen2.5-Coder-32B: 37.8 (#235), Qwen2-72B: 35.9 (#244)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| LMArena Non-English | 1205 | 1176 |
| LMArena Chinese | 1222 | 1240 |
| LMArena Russian | 1228 | 1169 |
| LMArena French | — | 1170 |
| LMArena German | — | 1151 |
| LMArena Japanese | — | 1111 |
| LMArena Korean | — | 1083 |
| LMArena Spanish | — | 1169 |
Instruction Following Too close to call
Qwen2.5-Coder-32B: 61.4 (#245), Qwen2-72B: 61.7 (#241)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| LMArena Instruction Following | 1223 | 1181 |
| LiveBench Instruction Following | 58.7% | — |
Long Context Qwen2.5-Coder-32B leads
Qwen2.5-Coder-32B: 38.0 (#208), Qwen2-72B: 36.1 (#235)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| LMArena Longer Query | 1251 | 1192 |
Writing & Preference Too close to call
Qwen2.5-Coder-32B: 41.6 (#240), Qwen2-72B: 40.8 (#241)
| Benchmark | Qwen2.5-Coder-32B | Qwen2-72B |
|---|---|---|
| LMArena Text | 1230 | 1203 |
| LMArena Creative Writing | 1174 | 1181 |
| LMArena Multi-Turn | 1222 | 1196 |
| LiveBench Language | 23.3% | — |
Frequently asked questions
Is Qwen2.5-Coder-32B better than Qwen2-72B?
Qwen2.5-Coder-32B is the stronger model overall, scoring 33.4 to 30.0 on the Noometry Index.
Is Qwen2.5-Coder-32B or Qwen2-72B better for coding?
Qwen2-72B scores higher on coding benchmarks: 29.1 versus 22.6 in the Noometry coding category.
How many benchmarks do Qwen2.5-Coder-32B and Qwen2-72B share?
16 benchmarks have published results for both models. Qwen2.5-Coder-32B has 31 scored results on Noometry and Qwen2-72B has 26.