Model comparison
DeepSeek-V3 vs Qwen2.5-Coder-32B
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 33.4 on the Noometry Index.
Last verified . 29 shared benchmarks.
DeepSeek
Alibaba (Qwen)Summary
- They share 29 benchmarks with published results for both. DeepSeek-V3 scores higher in 5 categories and Qwen2.5-Coder-32B in 3 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in coding, where DeepSeek-V3 leads 42.3 to 22.6.
- The biggest single-benchmark swing is Aider Polyglot: 55.1% for DeepSeek-V3 and 16.4% for Qwen2.5-Coder-32B.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $0.66 / $1 for Qwen2.5-Coder-32B.
- DeepSeek-V3 accepts more context: 164K tokens versus 33K.
Side by side
| DeepSeek-V3 | Qwen2.5-Coder-32B | |
|---|---|---|
| Provider | DeepSeek | Alibaba (Qwen) |
| Noometry Index | 39.5 | 33.4 |
| Released | 2024-12-26 | 2024-09-18 |
| Weights | Open | Open |
| Context window | 164K | 33K |
| Max output | 164K | 29K |
| Input $ / M tokens | $0.24 | $0.66 |
| Output $ / M tokens | $0.90 | $1 |
| Results tracked | 60 | 31 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3 leads
DeepSeek-V3: 42.3 (#106), Qwen2.5-Coder-32B: 22.6 (#333)
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| Aider Polyglot | 55.1% | 16.4% |
| BigCodeBench Instruct | 50% | 49% |
| LiveBench Coding | 70.9% | 56.9% |
| LMArena Coding | 1368 | 1276 |
| BigCodeBench Complete | 62.2% | 58% |
| HumanEval+ | 86.6% | 87.2% |
| MBPP+ | 73% | 77% |
| SWE-bench Verified (bash only) | — | 9% |
| SciCode | 35.8% | — |
| WeirdML | 36.1% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, Qwen2.5-Coder-32B: —
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| METR Time Horizons | 49.6% | — |
Reasoning Too close to call
DeepSeek-V3: 20.5 (#236), Qwen2.5-Coder-32B: 21.2 (#225)
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| LiveBench Reasoning | 65.8% | 42.1% |
| LMArena Hard Prompts | 1365 | 1251 |
| LiveBench Data Analysis | 60.9% | 49.9% |
| Epoch Capabilities Index | 135.94 | 119.49 |
| HellaSwag | 88.9% | 83% |
| LiveBench | 66.9% | 46.2% |
| WinoGrande | 85.2% | 80.8% |
| SimpleBench | 27.2% | — |
| Kagi LLM Benchmark | 52.3% | — |
| CritPt | 0% | — |
| DTBench | 64.8% | — |
| LMCA | 15.5% | — |
| BIG-Bench Hard | 87.5% | — |
| ForecastBench | 59.1 | — |
| PIQA | 84.7% | — |
Math Qwen2.5-Coder-32B leads
DeepSeek-V3: 32.1 (#219), Qwen2.5-Coder-32B: 33.3 (#204)
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| LiveBench Math | 73.5% | 46.6% |
| LMArena Math | 1373 | 1251 |
| OTIS Mock AIME 2024-2025 | 37.8% | — |
| Omni-MATH | 40.3% | — |
| MATH Level 5 | 75.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
| GSM8K | — | 93% |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#155), Qwen2.5-Coder-32B: 33.4 (#203)
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Expert | 1351 | 1221 |
| ARC (AI2) Challenge | 95.3% | 70.5% |
| MMLU | 87.2% | 79.1% |
| GPQA Diamond | 67.6% | — |
| MMLU-Pro | 72.3% | — |
| Confabulations | 26.1% | — |
| Vectara Hallucination Rate | 6.1% | — |
| GPQA (HELM) | 53.8% | — |
| TriviaQA | 82.9% | — |
Multilingual DeepSeek-V3 leads
DeepSeek-V3: 48.5 (#143), Qwen2.5-Coder-32B: 37.8 (#235)
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Non-English | 1358 | 1205 |
| LMArena Chinese | 1391 | 1222 |
| LMArena Russian | 1373 | 1228 |
| LMArena French | 1385 | — |
| LMArena German | 1374 | — |
| LMArena Japanese | 1333 | — |
| LMArena Korean | 1319 | — |
| LMArena Spanish | 1358 | — |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 72.8 (#130), Qwen2.5-Coder-32B: 61.4 (#245)
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| LiveBench Instruction Following | 81.5% | 58.7% |
| LMArena Instruction Following | 1345 | 1223 |
| IFEval | 83.2% | — |
Long Context Qwen2.5-Coder-32B leads
DeepSeek-V3: 34.0 (#253), Qwen2.5-Coder-32B: 38.0 (#208)
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Longer Query | 1352 | 1251 |
| Fiction.LiveBench | 50% | — |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.4 (#130), Qwen2.5-Coder-32B: 41.6 (#240)
| Benchmark | DeepSeek-V3 | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Text | 1375 | 1230 |
| LMArena Creative Writing | 1364 | 1174 |
| LMArena Multi-Turn | 1389 | 1222 |
| LiveBench Language | 49.1% | 23.3% |
| Short-Story Creative Writing | 77% | — |
| EQ-Bench Creative Writing | 1472 | — |
| WildBench | 83% | — |
Frequently asked questions
Is DeepSeek-V3 better than Qwen2.5-Coder-32B?
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 33.4 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or Qwen2.5-Coder-32B?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; Qwen2.5-Coder-32B lists at $0.66 and $1.
Is DeepSeek-V3 or Qwen2.5-Coder-32B better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 42.3 versus 22.6 in the Noometry coding category.
Which has the bigger context window?
DeepSeek-V3 does, with 164K tokens against 33K.
How many benchmarks do DeepSeek-V3 and Qwen2.5-Coder-32B share?
29 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and Qwen2.5-Coder-32B has 31.