Model comparison
Deepseek Coder v2 vs GPT-4
Deepseek Coder v2 is the stronger model overall, scoring 35.9 to 29.1 on the Noometry Index.
Last verified . 22 shared benchmarks.
DeepSeekSummary
- They share 22 benchmarks with published results for both. Deepseek Coder v2 scores higher in 5 categories and GPT-4 in 3 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where Deepseek Coder v2 leads 34.9 to 10.8.
- Deepseek Coder v2 has downloadable open weights; the other is API-only.
Side by side
| Deepseek Coder v2 | GPT-4 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 35.9 | 29.1 |
| Released | 2024-06-17 | 2023-03-14 |
| Weights | Open | Proprietary |
| Context window | — | 8K |
| Max output | — | 8K |
| Input $ / M tokens | — | $30 |
| Output $ / M tokens | — | $60 |
| Results tracked | 24 | 38 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Deepseek Coder v2 leads
Deepseek Coder v2: 38.1 (#183), GPT-4: 31.6 (#283)
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| BigCodeBench Instruct | 48.2% | 46% |
| LMArena Coding | 1251 | 1254 |
| BigCodeBench Complete | 59.7% | 57.2% |
| HumanEval+ | 82.3% | 79.3% |
| WeirdML | — | 12.4% |
| MBPP+ | 75.1% | — |
Agentic & Tool Use Not comparable
Deepseek Coder v2: —, GPT-4: —
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| METR Time Horizons | — | 36.1% |
Reasoning Deepseek Coder v2 leads
Deepseek Coder v2: 23.6 (#176), GPT-4: 17.8 (#289)
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| LMArena Hard Prompts | 1207 | 1241 |
| WinoGrande | 83.7% | 87.5% |
| Chess Puzzles | — | 4% |
| Mystery Game Puzzles | — | 12% |
| DTBench | — | 62.7% |
| LMCA | — | 17.1% |
| BIG-Bench Hard | — | 75.1% |
| Epoch Capabilities Index | — | 125.89 |
| ForecastBench | — | 57.8 |
| HellaSwag | — | 95.3% |
Math Deepseek Coder v2 leads
Deepseek Coder v2: 34.9 (#190), GPT-4: 10.8 (#309)
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| LMArena Math | 1241 | 1269 |
| GSM8K | 94.5% | 92% |
| OTIS Mock AIME 2024-2025 | — | 1.1% |
| MATH Level 5 | — | 23% |
Knowledge Deepseek Coder v2 leads
Deepseek Coder v2: 32.3 (#212), GPT-4: 18.4 (#282)
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| LMArena Expert | 1181 | 1211 |
| GPQA Diamond | — | 35.7% |
| ARC (AI2) Challenge | 64.3% | — |
| MMLU | — | 86.4% |
| TriviaQA | — | 84.8% |
Multilingual GPT-4 leads
Deepseek Coder v2: 36.3 (#240), GPT-4: 40.6 (#215)
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| LMArena Non-English | 1182 | 1246 |
| LMArena Chinese | 1201 | 1242 |
| LMArena French | 1185 | 1283 |
| LMArena German | 1164 | 1251 |
| LMArena Japanese | 1126 | 1209 |
| LMArena Korean | 1104 | 1184 |
| LMArena Russian | 1188 | 1251 |
| LMArena Spanish | 1153 | 1261 |
Instruction Following GPT-4 leads
Deepseek Coder v2: 61.7 (#242), GPT-4: 65.3 (#222)
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| LMArena Instruction Following | 1180 | 1241 |
Long Context Too close to call
Deepseek Coder v2: 37.0 (#224), GPT-4: 37.7 (#212)
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| LMArena Longer Query | 1219 | 1244 |
Writing & Preference Deepseek Coder v2 leads
Deepseek Coder v2: 38.2 (#253), GPT-4: 34.9 (#268)
| Benchmark | Deepseek Coder v2 | GPT-4 |
|---|---|---|
| LMArena Text | 1191 | 1263 |
| LMArena Creative Writing | 1120 | 1244 |
| LMArena Multi-Turn | 1177 | 1257 |
| EQ-Bench Creative Writing | — | 752 |
Frequently asked questions
Is Deepseek Coder v2 better than GPT-4?
Deepseek Coder v2 is the stronger model overall, scoring 35.9 to 29.1 on the Noometry Index.
Is Deepseek Coder v2 or GPT-4 better for coding?
Deepseek Coder v2 scores higher on coding benchmarks: 38.1 versus 31.6 in the Noometry coding category.
How many benchmarks do Deepseek Coder v2 and GPT-4 share?
22 benchmarks have published results for both models. Deepseek Coder v2 has 24 scored results on Noometry and GPT-4 has 38.