Model comparison
DeepSeek-V3 vs GPT-4 Turbo
DeepSeek-V3 is the stronger model overall, scoring 39.2 to 30.3 on the Noometry Index.
Last verified . 33 shared benchmarks.
DeepSeekSummary
- They share 33 benchmarks with published results for both. DeepSeek-V3 scores higher in 7 categories and GPT-4 Turbo in 1 category; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek-V3 leads 30.4 to 8.6.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 37.8% for DeepSeek-V3 and 6.7% for GPT-4 Turbo.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $10 / $30 for GPT-4 Turbo.
- DeepSeek-V3 accepts more context: 164K tokens versus 128K.
- DeepSeek-V3 has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V3 | GPT-4 Turbo | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 39.2 | 30.3 |
| Released | 2024-12-26 | 2023-11-06 |
| Weights | Open | Proprietary |
| Context window | 164K | 128K |
| Max output | 164K | 4K |
| Input $ / M tokens | $0.24 | $10 |
| Output $ / M tokens | $0.90 | $30 |
| Results tracked | 60 | 36 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3 leads
DeepSeek-V3: 41.9 (#99), GPT-4 Turbo: 33.4 (#247)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| WeirdML | 36.1% | 18% |
| BigCodeBench Instruct | 50% | 48.2% |
| LMArena Coding | 1368 | 1268 |
| BigCodeBench Complete | 62.2% | 58.2% |
| HumanEval+ | 86.6% | 86.6% |
| MBPP+ | 73% | 73.3% |
| Aider Polyglot | 55.1% | — |
| SciCode | 35.8% | — |
| LiveBench Coding | 70.9% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, GPT-4 Turbo: —
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| METR Time Horizons | 49.6% | 36.7% |
Reasoning DeepSeek-V3 leads
DeepSeek-V3: 20.0 (#226), GPT-4 Turbo: 14.9 (#311)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| SimpleBench | 27.2% | 25.1% |
| LMArena Hard Prompts | 1365 | 1251 |
| DTBench | 64.8% | 61.6% |
| LMCA | 15.5% | 9.8% |
| Epoch Capabilities Index | 135.95 | 127.25 |
| ForecastBench | 59.1 | 59.4 |
| Kagi LLM Benchmark | 52.3% | — |
| CritPt | 0% | — |
| Chess Puzzles | — | 6% |
| 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 leads
DeepSeek-V3: 30.4 (#208), GPT-4 Turbo: 8.6 (#316)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 6.7% |
| LMArena Math | 1373 | 1272 |
| MATH Level 5 | 75.5% | 46.7% |
| FrontierMath (Tiers 1-3) | — | 0.7% |
| Omni-MATH | 40.3% | — |
| LiveBench Math | 73.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#147), GPT-4 Turbo: 24.3 (#263)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| GPQA Diamond | 67.6% | 46.6% |
| Confabulations | 26.1% | 28.4% |
| LMArena Expert | 1352 | 1223 |
| MMLU | 87.2% | 81.3% |
| MMLU-Pro | 72.3% | — |
| Vectara Hallucination Rate | 6.1% | — |
| GPQA (HELM) | 53.8% | — |
| ARC (AI2) Challenge | 95.3% | — |
| TriviaQA | 82.9% | — |
Multimodal Not comparable
DeepSeek-V3: —, GPT-4 Turbo: 30.0 (#104)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| LMArena Vision | — | 1090 |
Multilingual DeepSeek-V3 leads
DeepSeek-V3: 48.8 (#138), GPT-4 Turbo: 40.7 (#211)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| LMArena Non-English | 1358 | 1245 |
| LMArena Chinese | 1390 | 1242 |
| LMArena French | 1386 | 1275 |
| LMArena German | 1374 | 1259 |
| LMArena Japanese | 1333 | 1194 |
| LMArena Korean | 1321 | 1187 |
| LMArena Russian | 1372 | 1259 |
| LMArena Spanish | 1358 | 1260 |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 73.0 (#125), GPT-4 Turbo: 66.0 (#211)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| LMArena Instruction Following | 1345 | 1250 |
| LiveBench Instruction Following | 81.5% | — |
| IFEval | 83.2% | — |
Long Context GPT-4 Turbo leads
DeepSeek-V3: 34.0 (#248), GPT-4 Turbo: 38.0 (#200)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| LMArena Longer Query | 1353 | 1254 |
| Fiction.LiveBench | 50% | — |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.6 (#124), GPT-4 Turbo: 47.9 (#201)
| Benchmark | DeepSeek-V3 | GPT-4 Turbo |
|---|---|---|
| LMArena Text | 1375 | 1272 |
| LMArena Creative Writing | 1364 | 1269 |
| LMArena Multi-Turn | 1389 | 1267 |
| Short-Story Creative Writing | 77% | — |
| EQ-Bench Creative Writing | 1472 | — |
| WildBench | 83% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than GPT-4 Turbo?
DeepSeek-V3 is the stronger model overall, scoring 39.2 to 30.3 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or GPT-4 Turbo?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; GPT-4 Turbo lists at $10 and $30.
Is DeepSeek-V3 or GPT-4 Turbo better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 41.9 versus 33.4 in the Noometry coding category.
Which has the bigger context window?
DeepSeek-V3 does, with 164K tokens against 128K.
How many benchmarks do DeepSeek-V3 and GPT-4 Turbo share?
33 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and GPT-4 Turbo has 36.