Model comparison
DeepSeek-V3.2-Speciale vs Qwen2.5 72B Instruct
DeepSeek-V3.2-Speciale is the stronger model overall, scoring 39.1 to 31.5 on the Noometry Index.
Last verified . 1 shared benchmarks.
DeepSeek
Alibaba (Qwen)Summary
- They share 1 benchmark with published results for both. DeepSeek-V3.2-Speciale scores higher in 2 categories and Qwen2.5 72B Instruct in 1 category; 2 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek-V3.2-Speciale leads 32.1 to 21.4.
- The biggest single-benchmark swing is WeirdML: 46.7% for DeepSeek-V3.2-Speciale and 16% for Qwen2.5 72B Instruct.
- DeepSeek-V3.2-Speciale is cheaper at $0.58 / $1.68 per million input/output tokens, against $1.40 / $5.60 for Qwen2.5 72B Instruct.
- Qwen2.5 72B Instruct accepts more context: 131K tokens versus 128K.
Side by side
| DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct | |
|---|---|---|
| Provider | DeepSeek | Alibaba (Qwen) |
| Noometry Index | 39.1 | 31.5 |
| Released | 2025-12-01 | 2024-09 |
| Weights | Open | Open |
| Context window | 128K | 131K |
| Max output | 128K | 8K |
| Input $ / M tokens | $0.58 | $1.40 |
| Output $ / M tokens | $1.68 | $5.60 |
| Results tracked | 3 | 43 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.2-Speciale leads
DeepSeek-V3.2-Speciale: 39.9 (#136), Qwen2.5 72B Instruct: 32.8 (#256)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| WeirdML | 46.7% | 16% |
| BigCodeBench Instruct | — | 45.8% |
| LMArena Coding | — | 1293 |
| BigCodeBench Complete | — | 55.9% |
Agentic & Tool Use Not comparable
DeepSeek-V3.2-Speciale: —, Qwen2.5 72B Instruct: 22.0 (#124)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| TheAgentCompany | — | 5.7% |
| BALROG | — | 16.2% |
| METR Time Horizons | — | 35.8% |
Reasoning DeepSeek-V3.2-Speciale leads
DeepSeek-V3.2-Speciale: 32.1 (#65), Qwen2.5 72B Instruct: 21.4 (#197)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| SimpleBench | 52.6% | — |
| LMArena Hard Prompts | — | 1271 |
| DTBench | — | 62.9% |
| LMCA | — | 13.4% |
| BIG-Bench Hard | — | 79.8% |
| Epoch Capabilities Index | — | 129.01 |
| ForecastBench | — | 57.5 |
| HellaSwag | — | 84.8% |
| PIQA | — | 82.6% |
| WinoGrande | — | 82.3% |
Math Not comparable
DeepSeek-V3.2-Speciale: —, Qwen2.5 72B Instruct: 18.1 (#281)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| OTIS Mock AIME 2024-2025 | — | 8.1% |
| Omni-MATH | — | 33% |
| LMArena Math | — | 1283 |
| MATH Level 5 | — | 63.2% |
Knowledge Not comparable
DeepSeek-V3.2-Speciale: —, Qwen2.5 72B Instruct: 26.9 (#247)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| GPQA Diamond | — | 49.1% |
| MMLU-Pro | — | 63.1% |
| Confabulations | — | 19.1% |
| GPQA (HELM) | — | 42.6% |
| LMArena Expert | — | 1245 |
| ARC (AI2) Challenge | — | 94.5% |
| MMLU | — | 85.3% |
| TriviaQA | — | 71.9% |
Multilingual Not comparable
DeepSeek-V3.2-Speciale: —, Qwen2.5 72B Instruct: 41.2 (#208)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Non-English | — | 1252 |
| LMArena Chinese | — | 1272 |
| LMArena French | — | 1279 |
| LMArena German | — | 1234 |
| LMArena Japanese | — | 1180 |
| LMArena Korean | — | 1188 |
| LMArena Russian | — | 1264 |
| LMArena Spanish | — | 1255 |
Instruction Following Not comparable
DeepSeek-V3.2-Speciale: —, Qwen2.5 72B Instruct: 65.7 (#216)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| IFEval | — | 80.6% |
| LMArena Instruction Following | — | 1254 |
Long Context Not comparable
DeepSeek-V3.2-Speciale: —, Qwen2.5 72B Instruct: 38.9 (#183)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Longer Query | — | 1282 |
Writing & Preference Too close to call
DeepSeek-V3.2-Speciale: 46.1 (#217), Qwen2.5 72B Instruct: 46.9 (#210)
| Benchmark | DeepSeek-V3.2-Speciale | Qwen2.5 72B Instruct |
|---|---|---|
| LMArena Text | — | 1269 |
| LMArena Creative Writing | — | 1221 |
| EQ-Bench Creative Writing | 1276 | — |
| WildBench | — | 80.2% |
| LMArena Multi-Turn | — | 1272 |
Frequently asked questions
Is DeepSeek-V3.2-Speciale better than Qwen2.5 72B Instruct?
DeepSeek-V3.2-Speciale is the stronger model overall, scoring 39.1 to 31.5 on the Noometry Index.
Which is cheaper, DeepSeek-V3.2-Speciale or Qwen2.5 72B Instruct?
DeepSeek-V3.2-Speciale is cheaper. It lists at $0.58 per million input tokens and $1.68 per million output tokens; Qwen2.5 72B Instruct lists at $1.40 and $5.60.
Is DeepSeek-V3.2-Speciale or Qwen2.5 72B Instruct better for coding?
DeepSeek-V3.2-Speciale scores higher on coding benchmarks: 39.9 versus 32.8 in the Noometry coding category.
Which has the bigger context window?
Qwen2.5 72B Instruct does, with 131K tokens against 128K.
How many benchmarks do DeepSeek-V3.2-Speciale and Qwen2.5 72B Instruct share?
1 benchmark has published results for both models. DeepSeek-V3.2-Speciale has 3 scored results on Noometry and Qwen2.5 72B Instruct has 43.