Model comparison
DeepSeek-V3 vs DeepSeek-V3.1
DeepSeek-V3.1 is the stronger model overall, scoring 42.8 to 39.5 on the Noometry Index.
Last verified . 27 shared benchmarks.
DeepSeekSummary
- They share 27 benchmarks with published results for both. DeepSeek-V3 scores higher in 1 category and DeepSeek-V3.1 in 7 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek-V3.1 leads 27.9 to 20.5.
- The biggest single-benchmark swing is DTBench: 64.8% for DeepSeek-V3 and 82.7% for DeepSeek-V3.1.
- Both cost about the same: $0.24 input and $0.90 output per million tokens.
Side by side
| DeepSeek-V3 | DeepSeek-V3.1 | |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 39.5 | 42.8 |
| Released | 2024-12-26 | 2025-08-21 |
| Weights | Open | Open |
| Context window | 164K | 164K |
| Max output | 164K | 8K |
| Input $ / M tokens | $0.24 | $0.25 |
| Output $ / M tokens | $0.90 | $0.95 |
| Results tracked | 60 | 27 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3 leads
DeepSeek-V3: 42.3 (#106), DeepSeek-V3.1: 40.3 (#144)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| WeirdML | 36.1% | 38.4% |
| LMArena Coding | 1368 | 1417 |
| Aider Polyglot | 55.1% | — |
| SciCode | 35.8% | — |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, DeepSeek-V3.1: —
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| METR Time Horizons | 49.6% | — |
Reasoning DeepSeek-V3.1 leads
DeepSeek-V3: 20.5 (#236), DeepSeek-V3.1: 27.9 (#110)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| SimpleBench | 27.2% | 40% |
| Kagi LLM Benchmark | 52.3% | 53.2% |
| LMArena Hard Prompts | 1365 | 1417 |
| DTBench | 64.8% | 82.7% |
| LMCA | 15.5% | 24.3% |
| Epoch Capabilities Index | 135.94 | 139.92 |
| ForecastBench | 59.1 | 58 |
| CritPt | 0% | — |
| LiveBench Reasoning | 65.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| BIG-Bench Hard | 87.5% | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math DeepSeek-V3.1 leads
DeepSeek-V3: 32.1 (#219), DeepSeek-V3.1: 38.9 (#122)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| LMArena Math | 1373 | 1420 |
| OTIS Mock AIME 2024-2025 | 37.8% | — |
| Omni-MATH | 40.3% | — |
| LiveBench Math | 73.5% | — |
| MATH Level 5 | 75.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
Knowledge DeepSeek-V3.1 leads
DeepSeek-V3: 37.5 (#155), DeepSeek-V3.1: 43.7 (#90)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| Vectara Hallucination Rate | 6.1% | 5.5% |
| LMArena Expert | 1351 | 1405 |
| GPQA Diamond | 67.6% | — |
| MMLU-Pro | 72.3% | — |
| Confabulations | 26.1% | — |
| GPQA (HELM) | 53.8% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |
Multilingual DeepSeek-V3.1 leads
DeepSeek-V3: 48.5 (#143), DeepSeek-V3.1: 51.6 (#106)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| LMArena Non-English | 1358 | 1400 |
| LMArena Chinese | 1391 | 1469 |
| LMArena French | 1385 | 1447 |
| LMArena German | 1374 | 1411 |
| LMArena Japanese | 1333 | 1378 |
| LMArena Korean | 1319 | 1337 |
| LMArena Russian | 1373 | 1405 |
| LMArena Spanish | 1358 | 1431 |
Instruction Following DeepSeek-V3.1 leads
DeepSeek-V3: 72.8 (#130), DeepSeek-V3.1: 73.9 (#110)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| LMArena Instruction Following | 1345 | 1400 |
| LiveBench Instruction Following | 81.5% | — |
| IFEval | 83.2% | — |
Long Context DeepSeek-V3.1 leads
DeepSeek-V3: 34.0 (#253), DeepSeek-V3.1: 36.3 (#232)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| Fiction.LiveBench | 50% | 52.8% |
| LMArena Longer Query | 1352 | 1422 |
Writing & Preference DeepSeek-V3.1 leads
DeepSeek-V3: 57.4 (#130), DeepSeek-V3.1: 60.3 (#98)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.1 |
|---|---|---|
| LMArena Text | 1375 | 1420 |
| LMArena Creative Writing | 1364 | 1401 |
| EQ-Bench Creative Writing | 1472 | 1436 |
| LMArena Multi-Turn | 1389 | 1408 |
| Short-Story Creative Writing | 77% | — |
| WildBench | 83% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than DeepSeek-V3.1?
DeepSeek-V3.1 is the stronger model overall, scoring 42.8 to 39.5 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or DeepSeek-V3.1?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; DeepSeek-V3.1 lists at $0.25 and $0.95.
Is DeepSeek-V3 or DeepSeek-V3.1 better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 42.3 versus 40.3 in the Noometry coding category.
Which has the bigger context window?
Both accept 164K tokens.
How many benchmarks do DeepSeek-V3 and DeepSeek-V3.1 share?
27 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and DeepSeek-V3.1 has 27.