Model comparison
DeepSeek-V2 (MoE-236B, May 2024) vs GPT-4
GPT-4 has enough public results to be ranked (#316); DeepSeek-V2 (MoE-236B, May 2024) does not yet, so treat this comparison as directional.
Last verified . 8 shared benchmarks.
DeepSeekSummary
- They share 8 benchmarks with published results for both. DeepSeek-V2 (MoE-236B, May 2024) scores higher in 1 category and GPT-4 in 0 categories; one gap is clear of the uncertainty.
- The widest gap is in coding, where DeepSeek-V2 (MoE-236B, May 2024) leads 40.4 to 31.6.
- DeepSeek-V2 (MoE-236B, May 2024) has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 40.3 | 29.1 |
| Released | 2024-05-07 | 2023-03-14 |
| Weights | Open | Proprietary |
| Context window | — | 8K |
| Max output | — | 8K |
| Input $ / M tokens | — | $30 |
| Output $ / M tokens | — | $60 |
| Results tracked | 10 | 38 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V2 (MoE-236B, May 2024) leads
DeepSeek-V2 (MoE-236B, May 2024): 40.4 (#139), GPT-4: 31.6 (#283)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| BigCodeBench Instruct | 48.9% | 46% |
| BigCodeBench Complete | 59.4% | 57.2% |
| WeirdML | — | 12.4% |
| LMArena Coding | — | 1254 |
| HumanEval+ | — | 79.3% |
Agentic & Tool Use Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, GPT-4: —
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| METR Time Horizons | — | 36.1% |
Reasoning Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, GPT-4: 17.8 (#289)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| BIG-Bench Hard | 78.8% | 75.1% |
| Epoch Capabilities Index | 124.77 | 125.89 |
| HellaSwag | 87.1% | 95.3% |
| WinoGrande | 86.3% | 87.5% |
| Chess Puzzles | — | 4% |
| LMArena Hard Prompts | — | 1241 |
| Mystery Game Puzzles | — | 12% |
| DTBench | — | 62.7% |
| LMCA | — | 17.1% |
| ForecastBench | — | 57.8 |
| PIQA | 83.9% | — |
Math Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, GPT-4: 10.8 (#309)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | — | 1.1% |
| LMArena Math | — | 1269 |
| MATH Level 5 | — | 23% |
| GSM8K | — | 92% |
Knowledge Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, GPT-4: 18.4 (#282)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| MMLU | 78.4% | 86.4% |
| TriviaQA | 80% | 84.8% |
| GPQA Diamond | — | 35.7% |
| LMArena Expert | — | 1211 |
| ARC (AI2) Challenge | 92.2% | — |
Multilingual Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, GPT-4: 40.6 (#215)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| LMArena Non-English | — | 1246 |
| LMArena Chinese | — | 1242 |
| LMArena French | — | 1283 |
| LMArena German | — | 1251 |
| LMArena Japanese | — | 1209 |
| LMArena Korean | — | 1184 |
| LMArena Russian | — | 1251 |
| LMArena Spanish | — | 1261 |
Instruction Following Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, GPT-4: 65.3 (#222)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| LMArena Instruction Following | — | 1241 |
Long Context Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, GPT-4: 37.7 (#212)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| LMArena Longer Query | — | 1244 |
Writing & Preference Not comparable
DeepSeek-V2 (MoE-236B, May 2024): —, GPT-4: 34.9 (#268)
| Benchmark | DeepSeek-V2 (MoE-236B, May 2024) | GPT-4 |
|---|---|---|
| LMArena Text | — | 1263 |
| LMArena Creative Writing | — | 1244 |
| EQ-Bench Creative Writing | — | 752 |
| LMArena Multi-Turn | — | 1257 |
Frequently asked questions
Is DeepSeek-V2 (MoE-236B, May 2024) better than GPT-4?
GPT-4 has enough public results to be ranked (#316); DeepSeek-V2 (MoE-236B, May 2024) does not yet, so treat this comparison as directional.
Is DeepSeek-V2 (MoE-236B, May 2024) or GPT-4 better for coding?
DeepSeek-V2 (MoE-236B, May 2024) scores higher on coding benchmarks: 40.4 versus 31.6 in the Noometry coding category.
How many benchmarks do DeepSeek-V2 (MoE-236B, May 2024) and GPT-4 share?
8 benchmarks have published results for both models. DeepSeek-V2 (MoE-236B, May 2024) has 10 scored results on Noometry and GPT-4 has 38.