Model comparison
DeepSeek-V3.1 vs DeepSeek-V3.2-Speciale
DeepSeek-V3.1 is the stronger model overall, scoring 42.2 to 39.1 on the Noometry Index.
Last verified . 3 shared benchmarks.
DeepSeekSummary
- They share 3 benchmarks with published results for both. DeepSeek-V3.1 scores higher in 1 category and DeepSeek-V3.2-Speciale in 2 categories; 2 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where DeepSeek-V3.1 leads 60.5 to 46.1.
- The biggest single-benchmark swing is SimpleBench: 40% for DeepSeek-V3.1 and 52.6% for DeepSeek-V3.2-Speciale.
- DeepSeek-V3.1 is cheaper at $0.25 / $0.95 per million input/output tokens, against $0.58 / $1.68 for DeepSeek-V3.2-Speciale.
- DeepSeek-V3.1 accepts more context: 164K tokens versus 128K.
Side by side
| DeepSeek-V3.1 | DeepSeek-V3.2-Speciale | |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 42.2 | 39.1 |
| Released | 2025-08-21 | 2025-12-01 |
| Weights | Open | Open |
| Context window | 164K | 128K |
| Max output | 8K | 128K |
| Input $ / M tokens | $0.25 | $0.58 |
| Output $ / M tokens | $0.95 | $1.68 |
| Results tracked | 27 | 3 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
DeepSeek-V3.1: 39.8 (#140), DeepSeek-V3.2-Speciale: 39.9 (#136)
| Benchmark | DeepSeek-V3.1 | DeepSeek-V3.2-Speciale |
|---|---|---|
| WeirdML | 38.4% | 46.7% |
| LMArena Coding | 1417 | — |
Reasoning DeepSeek-V3.2-Speciale leads
DeepSeek-V3.1: 27.3 (#109), DeepSeek-V3.2-Speciale: 32.1 (#65)
| Benchmark | DeepSeek-V3.1 | DeepSeek-V3.2-Speciale |
|---|---|---|
| SimpleBench | 40% | 52.6% |
| Kagi LLM Benchmark | 53.2% | — |
| LMArena Hard Prompts | 1417 | — |
| DTBench | 82.7% | — |
| LMCA | 24.3% | — |
| Epoch Capabilities Index | 139.92 | — |
| ForecastBench | 58 | — |
Math Not comparable
DeepSeek-V3.1: 36.1 (#125), DeepSeek-V3.2-Speciale: —
| Benchmark | DeepSeek-V3.1 | DeepSeek-V3.2-Speciale |
|---|---|---|
| LMArena Math | 1420 | — |
Knowledge Not comparable
DeepSeek-V3.1: 43.0 (#86), DeepSeek-V3.2-Speciale: —
| Benchmark | DeepSeek-V3.1 | DeepSeek-V3.2-Speciale |
|---|---|---|
| Vectara Hallucination Rate | 5.5% | — |
| LMArena Expert | 1403 | — |
Multilingual Not comparable
DeepSeek-V3.1: 51.8 (#100), DeepSeek-V3.2-Speciale: —
| Benchmark | DeepSeek-V3.1 | DeepSeek-V3.2-Speciale |
|---|---|---|
| LMArena Non-English | 1400 | — |
| LMArena Chinese | 1470 | — |
| LMArena French | 1447 | — |
| LMArena German | 1412 | — |
| LMArena Japanese | 1374 | — |
| LMArena Korean | 1337 | — |
| LMArena Russian | 1405 | — |
| LMArena Spanish | 1429 | — |
Instruction Following Not comparable
DeepSeek-V3.1: 74.1 (#105), DeepSeek-V3.2-Speciale: —
| Benchmark | DeepSeek-V3.1 | DeepSeek-V3.2-Speciale |
|---|---|---|
| LMArena Instruction Following | 1400 | — |
Long Context Not comparable
DeepSeek-V3.1: 36.3 (#227), DeepSeek-V3.2-Speciale: —
| Benchmark | DeepSeek-V3.1 | DeepSeek-V3.2-Speciale |
|---|---|---|
| Fiction.LiveBench | 52.8% | — |
| LMArena Longer Query | 1422 | — |
Writing & Preference DeepSeek-V3.1 leads
DeepSeek-V3.1: 60.5 (#93), DeepSeek-V3.2-Speciale: 46.1 (#217)
| Benchmark | DeepSeek-V3.1 | DeepSeek-V3.2-Speciale |
|---|---|---|
| EQ-Bench Creative Writing | 1436 | 1276 |
| LMArena Text | 1420 | — |
| LMArena Creative Writing | 1401 | — |
| LMArena Multi-Turn | 1408 | — |
Frequently asked questions
Is DeepSeek-V3.1 better than DeepSeek-V3.2-Speciale?
DeepSeek-V3.1 is the stronger model overall, scoring 42.2 to 39.1 on the Noometry Index.
Which is cheaper, DeepSeek-V3.1 or DeepSeek-V3.2-Speciale?
DeepSeek-V3.1 is cheaper. It lists at $0.25 per million input tokens and $0.95 per million output tokens; DeepSeek-V3.2-Speciale lists at $0.58 and $1.68.
Is DeepSeek-V3.1 or DeepSeek-V3.2-Speciale better for coding?
They score almost the same on coding (39.8 vs 39.9); test both on your own repository before choosing.
Which has the bigger context window?
DeepSeek-V3.1 does, with 164K tokens against 128K.
How many benchmarks do DeepSeek-V3.1 and DeepSeek-V3.2-Speciale share?
3 benchmarks have published results for both models. DeepSeek-V3.1 has 27 scored results on Noometry and DeepSeek-V3.2-Speciale has 3.