Model comparison
DeepSeek-V3 vs GPT-4o
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 28.6 on the Noometry Index.
Last verified . 51 shared benchmarks.
DeepSeekSummary
- They share 51 benchmarks with published results for both. DeepSeek-V3 scores higher in 7 categories and GPT-4o in 1 category; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek-V3 leads 32.1 to 10.6.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 37.8% for DeepSeek-V3 and 6.4% for GPT-4o.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $2.50 / $10 for GPT-4o.
- 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-4o | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 39.5 | 28.6 |
| Released | 2024-12-26 | 2024-05-13 |
| Weights | Open | Proprietary |
| Context window | 164K | 128K |
| Max output | 164K | 16K |
| Input $ / M tokens | $0.24 | $2.50 |
| Output $ / M tokens | $0.90 | $10 |
| Results tracked | 60 | 72 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3 leads
DeepSeek-V3: 42.3 (#106), GPT-4o: 24.8 (#328)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| Aider Polyglot | 55.1% | 45.3% |
| WeirdML | 36.1% | 25.1% |
| BigCodeBench Instruct | 50% | 51.1% |
| LiveBench Coding | 70.9% | 51.4% |
| LMArena Coding | 1368 | 1297 |
| BigCodeBench Complete | 62.2% | 61.1% |
| HumanEval+ | 86.6% | 87.2% |
| MBPP+ | 73% | 72.2% |
| SWE-bench Verified | — | 31% |
| SWE-bench Verified (bash only) | — | 21.6% |
| SciCode | 35.8% | — |
| GSO | — | 0% |
| CadEval | — | 26% |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, GPT-4o: 21.0 (#141)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| METR Time Horizons | 49.6% | 40.8% |
| GDPval | — | 9.9% |
| TheAgentCompany | — | 8.6% |
| Cybench | — | 12.5% |
| BALROG | — | 32.3% |
| LMArena Search | — | 1006 |
Reasoning DeepSeek-V3 leads
DeepSeek-V3: 20.5 (#236), GPT-4o: 9.4 (#343)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| SimpleBench | 27.2% | 17.8% |
| CritPt | 0% | 0% |
| LiveBench Reasoning | 65.8% | 55.8% |
| LMArena Hard Prompts | 1365 | 1281 |
| DTBench | 64.8% | 64.5% |
| LiveBench Data Analysis | 60.9% | 60.9% |
| LMCA | 15.5% | 16.6% |
| Epoch Capabilities Index | 135.94 | 128.97 |
| ForecastBench | 59.1 | 57.7 |
| LiveBench | 66.9% | 55.3% |
| ARC-AGI-2 | — | 0% |
| Kagi LLM Benchmark | 52.3% | — |
| ARC-AGI-1 | — | 4.5% |
| Chess Puzzles | — | 13% |
| EnigmaEval | — | 0.8% |
| BIG-Bench Hard | 87.5% | — |
| HellaSwag | 88.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math DeepSeek-V3 leads
DeepSeek-V3: 32.1 (#219), GPT-4o: 10.6 (#312)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 6.4% |
| Omni-MATH | 40.3% | 29.3% |
| LiveBench Math | 73.5% | 49.5% |
| LMArena Math | 1373 | 1285 |
| MATH Level 5 | 75.5% | 53.3% |
| FrontierMath (Feb 2025 set) | 1.7% | 0.3% |
| FrontierMath (Tiers 1-3) | — | 0.4% |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#155), GPT-4o: 28.8 (#242)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| GPQA Diamond | 67.6% | 49.2% |
| MMLU-Pro | 72.3% | 71.3% |
| Confabulations | 26.1% | 15.3% |
| Vectara Hallucination Rate | 6.1% | 9.6% |
| GPQA (HELM) | 53.8% | 52% |
| LMArena Expert | 1351 | 1250 |
| MMLU | 87.2% | 88.1% |
| Humanity's Last Exam | — | 2.7% |
| SimpleQA Verified | — | 26% |
| ARC (AI2) Challenge | 95.3% | — |
| TriviaQA | 82.9% | — |
Multimodal Not comparable
DeepSeek-V3: —, GPT-4o: 34.5 (#91)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| LMArena Vision | — | 1137 |
| Video-MME | — | 71.9% |
| GeoBench | — | 71% |
| VPCT | — | 40% |
| ScienceQA | — | 88.5% |
Multilingual DeepSeek-V3 leads
DeepSeek-V3: 48.5 (#143), GPT-4o: 43.2 (#186)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| LMArena Non-English | 1358 | 1283 |
| LMArena Chinese | 1391 | 1277 |
| LMArena French | 1385 | 1304 |
| LMArena German | 1374 | 1282 |
| LMArena Japanese | 1333 | 1257 |
| LMArena Korean | 1319 | 1234 |
| LMArena Russian | 1373 | 1286 |
| LMArena Spanish | 1358 | 1292 |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 72.8 (#130), GPT-4o: 66.6 (#207)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| LiveBench Instruction Following | 81.5% | 68.6% |
| IFEval | 83.2% | 81.7% |
| LMArena Instruction Following | 1345 | 1278 |
Long Context GPT-4o leads
DeepSeek-V3: 34.0 (#253), GPT-4o: 39.4 (#179)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| Fiction.LiveBench | 50% | 66.7% |
| LMArena Longer Query | 1352 | 1289 |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.4 (#130), GPT-4o: 52.6 (#166)
| Benchmark | DeepSeek-V3 | GPT-4o |
|---|---|---|
| LMArena Text | 1375 | 1300 |
| LMArena Creative Writing | 1364 | 1292 |
| Short-Story Creative Writing | 77% | 81.8% |
| WildBench | 83% | 82.8% |
| LMArena Multi-Turn | 1389 | 1302 |
| LiveBench Language | 49.1% | 47.6% |
| EQ-Bench Creative Writing | 1472 | — |
Frequently asked questions
Is DeepSeek-V3 better than GPT-4o?
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 28.6 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or GPT-4o?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; GPT-4o lists at $2.50 and $10.
Is DeepSeek-V3 or GPT-4o better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 42.3 versus 24.8 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-4o share?
51 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and GPT-4o has 72.