Model comparison
DeepSeek-V2.5 (Sep 2024) vs DeepSeek-V3.1
DeepSeek-V3.1 is the stronger model overall, scoring 42.8 to 37.6 on the Noometry Index.
Last verified . 17 shared benchmarks.
DeepSeekSummary
- They share 17 benchmarks with published results for both. DeepSeek-V2.5 (Sep 2024) scores higher in 1 category and DeepSeek-V3.1 in 7 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where DeepSeek-V3.1 leads 60.3 to 49.8.
Side by side
| DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 | |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 37.6 | 42.8 |
| Released | 2024-09-06 | 2025-08-21 |
| Weights | Open | Open |
| Context window | — | 164K |
| Max output | — | 8K |
| Input $ / M tokens | — | $0.25 |
| Output $ / M tokens | — | $0.95 |
| Results tracked | 22 | 27 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.1 leads
DeepSeek-V2.5 (Sep 2024): 31.7 (#281), DeepSeek-V3.1: 40.3 (#144)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 |
|---|---|---|
| LMArena Coding | 1309 | 1417 |
| Aider Polyglot | 17.8% | — |
| WeirdML | — | 38.4% |
| BigCodeBench Instruct | 48.6% | — |
| BigCodeBench Complete | 53.2% | — |
| HumanEval+ | 83.5% | — |
| MBPP+ | 74.1% | — |
Reasoning DeepSeek-V3.1 leads
DeepSeek-V2.5 (Sep 2024): 25.6 (#145), DeepSeek-V3.1: 27.9 (#110)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 |
|---|---|---|
| LMArena Hard Prompts | 1289 | 1417 |
| SimpleBench | — | 40% |
| Kagi LLM Benchmark | — | 53.2% |
| DTBench | — | 82.7% |
| LMCA | — | 24.3% |
| Epoch Capabilities Index | — | 139.92 |
| ForecastBench | — | 58 |
Math DeepSeek-V3.1 leads
DeepSeek-V2.5 (Sep 2024): 35.9 (#177), DeepSeek-V3.1: 38.9 (#122)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 |
|---|---|---|
| LMArena Math | 1288 | 1420 |
Knowledge DeepSeek-V3.1 leads
DeepSeek-V2.5 (Sep 2024): 34.8 (#193), DeepSeek-V3.1: 43.7 (#90)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 |
|---|---|---|
| LMArena Expert | 1266 | 1405 |
| Vectara Hallucination Rate | — | 5.5% |
Multilingual DeepSeek-V3.1 leads
DeepSeek-V2.5 (Sep 2024): 42.5 (#193), DeepSeek-V3.1: 51.6 (#106)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 |
|---|---|---|
| LMArena Non-English | 1273 | 1400 |
| LMArena Chinese | 1318 | 1469 |
| LMArena French | 1289 | 1447 |
| LMArena German | 1258 | 1411 |
| LMArena Japanese | 1228 | 1378 |
| LMArena Korean | 1209 | 1337 |
| LMArena Russian | 1289 | 1405 |
| LMArena Spanish | 1248 | 1431 |
Instruction Following DeepSeek-V3.1 leads
DeepSeek-V2.5 (Sep 2024): 67.5 (#194), DeepSeek-V3.1: 73.9 (#110)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 |
|---|---|---|
| LMArena Instruction Following | 1280 | 1400 |
Long Context DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 39.5 (#174), DeepSeek-V3.1: 36.3 (#232)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 |
|---|---|---|
| LMArena Longer Query | 1301 | 1422 |
| Fiction.LiveBench | — | 52.8% |
Writing & Preference DeepSeek-V3.1 leads
DeepSeek-V2.5 (Sep 2024): 49.8 (#187), DeepSeek-V3.1: 60.3 (#98)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek-V3.1 |
|---|---|---|
| LMArena Text | 1294 | 1420 |
| LMArena Creative Writing | 1285 | 1401 |
| LMArena Multi-Turn | 1297 | 1408 |
| EQ-Bench Creative Writing | — | 1436 |
Frequently asked questions
Is DeepSeek-V2.5 (Sep 2024) better than DeepSeek-V3.1?
DeepSeek-V3.1 is the stronger model overall, scoring 42.8 to 37.6 on the Noometry Index.
Is DeepSeek-V2.5 (Sep 2024) or DeepSeek-V3.1 better for coding?
DeepSeek-V3.1 scores higher on coding benchmarks: 40.3 versus 31.7 in the Noometry coding category.
How many benchmarks do DeepSeek-V2.5 (Sep 2024) and DeepSeek-V3.1 share?
17 benchmarks have published results for both models. DeepSeek-V2.5 (Sep 2024) has 22 scored results on Noometry and DeepSeek-V3.1 has 27.