Model comparison
DeepSeek-V2 (MoE-236B, May 2024) vs DeepSeek-V3
DeepSeek-V3 has enough public results to be ranked (#166); DeepSeek-V2 (MoE-236B, May 2024) does not yet, so treat this comparison as directional.
Last verified . 10 shared benchmarks.
DeepSeekSummary
- They share 10 benchmarks with published results for both. DeepSeek-V2 (MoE-236B, May 2024) scores higher in 0 categories and DeepSeek-V3 in 1 category; one gap is clear of the uncertainty.
Side by side
| DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 | |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 40.3 | 39.5 |
| Released | 2024-05-07 | 2024-12-26 |
| Weights | Open | Open |
| Context window | — | 164K |
| Max output | — | 164K |
| Input $ / M tokens | — | $0.24 |
| Output $ / M tokens | — | $0.90 |
| Results tracked | 10 | 60 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3 leads
DeepSeek-V2 (MoE-236B, May 2024): 40.4 (#139), DeepSeek-V3: 42.3 (#106)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| BigCodeBench Instruct | 48.9% | 50% |
| BigCodeBench Complete | 59.4% | 62.2% |
| Aider Polyglot | — | 55.1% |
| SciCode | — | 35.8% |
| WeirdML | — | 36.1% |
| LiveBench Coding | — | 70.9% |
| LMArena Coding | — | 1368 |
| HumanEval+ | — | 86.6% |
| MBPP+ | — | 73% |
Agentic & Tool Use Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, DeepSeek-V3: —
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| METR Time Horizons | — | 49.6% |
Reasoning Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, DeepSeek-V3: 20.5 (#236)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| BIG-Bench Hard | 78.8% | 87.5% |
| Epoch Capabilities Index | 124.77 | 135.94 |
| HellaSwag | 87.1% | 88.9% |
| PIQA | 83.9% | 84.7% |
| WinoGrande | 86.3% | 85.2% |
| SimpleBench | — | 27.2% |
| Kagi LLM Benchmark | — | 52.3% |
| CritPt | — | 0% |
| LiveBench Reasoning | — | 65.8% |
| LMArena Hard Prompts | — | 1365 |
| DTBench | — | 64.8% |
| LiveBench Data Analysis | — | 60.9% |
| LMCA | — | 15.5% |
| ForecastBench | — | 59.1 |
| LiveBench | — | 66.9% |
Math Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, DeepSeek-V3: 32.1 (#219)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | — | 37.8% |
| Omni-MATH | — | 40.3% |
| LiveBench Math | — | 73.5% |
| LMArena Math | — | 1373 |
| MATH Level 5 | — | 75.5% |
| FrontierMath (Feb 2025 set) | — | 1.7% |
Knowledge Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, DeepSeek-V3: 37.5 (#155)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| ARC (AI2) Challenge | 92.2% | 95.3% |
| MMLU | 78.4% | 87.2% |
| TriviaQA | 80% | 82.9% |
| GPQA Diamond | — | 67.6% |
| MMLU-Pro | — | 72.3% |
| Confabulations | — | 26.1% |
| Vectara Hallucination Rate | — | 6.1% |
| GPQA (HELM) | — | 53.8% |
| LMArena Expert | — | 1351 |
Multilingual Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, DeepSeek-V3: 48.5 (#143)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| LMArena Non-English | — | 1358 |
| LMArena Chinese | — | 1391 |
| LMArena French | — | 1385 |
| LMArena German | — | 1374 |
| LMArena Japanese | — | 1333 |
| LMArena Korean | — | 1319 |
| LMArena Russian | — | 1373 |
| LMArena Spanish | — | 1358 |
Instruction Following Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, DeepSeek-V3: 72.8 (#130)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| LiveBench Instruction Following | — | 81.5% |
| IFEval | — | 83.2% |
| LMArena Instruction Following | — | 1345 |
Long Context Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, DeepSeek-V3: 34.0 (#253)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| Fiction.LiveBench | — | 50% |
| LMArena Longer Query | — | 1352 |
Writing & Preference Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, DeepSeek-V3: 57.4 (#130)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | DeepSeek-V3 |
|---|---|---|
| LMArena Text | — | 1375 |
| LMArena Creative Writing | — | 1364 |
| Short-Story Creative Writing | — | 77% |
| EQ-Bench Creative Writing | — | 1472 |
| WildBench | — | 83% |
| LMArena Multi-Turn | — | 1389 |
| LiveBench Language | — | 49.1% |
Frequently asked questions
Is DeepSeek-V2 (MoE-236B, May 2024) better than DeepSeek-V3?
DeepSeek-V3 has enough public results to be ranked (#166); DeepSeek-V2 (MoE-236B, May 2024) does not yet, so treat this comparison as directional.
Is DeepSeek-V2 (MoE-236B, May 2024) or DeepSeek-V3 better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 42.3 versus 40.4 in the Noometry coding category.
How many benchmarks do DeepSeek-V2 (MoE-236B, May 2024) and DeepSeek-V3 share?
10 benchmarks have published results for both models. DeepSeek-V2 (MoE-236B, May 2024) has 10 scored results on Noometry and DeepSeek-V3 has 60.