Model comparison
DeepSeek-V3 vs GPT-4.5
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 37.2 on the Noometry Index.
Last verified . 35 shared benchmarks.
DeepSeekSummary
- They share 35 benchmarks with published results for both. DeepSeek-V3 scores higher in 5 categories and GPT-4.5 in 3 categories; 4 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek-V3 leads 20.5 to 13.9.
- The biggest single-benchmark swing is Fiction.LiveBench: 50% for DeepSeek-V3 and 63.9% for GPT-4.5.
- DeepSeek-V3 has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V3 | GPT-4.5 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 39.5 | 37.2 |
| Released | 2024-12-26 | 2025-02-27 |
| Weights | Open | Proprietary |
| Context window | 164K | — |
| Max output | 164K | — |
| Input $ / M tokens | $0.24 | — |
| Output $ / M tokens | $0.90 | — |
| Results tracked | 60 | 42 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
DeepSeek-V3: 42.3 (#106), GPT-4.5: 42.2 (#109)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| Aider Polyglot | 55.1% | 44.9% |
| WeirdML | 36.1% | 39.4% |
| LiveBench Coding | 70.9% | 75.2% |
| LMArena Coding | 1368 | 1396 |
| SciCode | 35.8% | — |
| BigCodeBench Instruct | 50% | — |
| BigCodeBench Complete | 62.2% | — |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, GPT-4.5: 27.9 (#97)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| Cybench | — | 17.5% |
| METR Time Horizons | 49.6% | — |
Reasoning DeepSeek-V3 leads
DeepSeek-V3: 20.5 (#236), GPT-4.5: 13.9 (#330)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| SimpleBench | 27.2% | 34.5% |
| LiveBench Reasoning | 65.8% | 71.1% |
| LMArena Hard Prompts | 1365 | 1403 |
| LiveBench Data Analysis | 60.9% | 64.3% |
| Epoch Capabilities Index | 135.94 | 136.74 |
| ForecastBench | 59.1 | 61.7 |
| LiveBench | 66.9% | 69% |
| ARC-AGI-2 | — | 0.8% |
| Kagi LLM Benchmark | 52.3% | — |
| ARC-AGI-1 | — | 10.3% |
| CritPt | 0% | — |
| EnigmaEval | — | 3.2% |
| DTBench | 64.8% | — |
| LMCA | 15.5% | — |
| BIG-Bench Hard | 87.5% | — |
| HellaSwag | 88.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math Too close to call
DeepSeek-V3: 32.1 (#219), GPT-4.5: 32.6 (#211)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 37.8% |
| LiveBench Math | 73.5% | 69.3% |
| LMArena Math | 1373 | 1412 |
| MATH Level 5 | 75.5% | 78.6% |
| Omni-MATH | 40.3% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#155), GPT-4.5: 32.5 (#211)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| GPQA Diamond | 67.6% | 68.7% |
| Confabulations | 26.1% | 13.6% |
| LMArena Expert | 1351 | 1394 |
| Humanity's Last Exam | — | 5.4% |
| MMLU-Pro | 72.3% | — |
| Vectara Hallucination Rate | 6.1% | — |
| GPQA (HELM) | 53.8% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |
Multimodal Not comparable
DeepSeek-V3: —, GPT-4.5: 37.6 (#71)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| LMArena Vision | — | 1195 |
| VPCT | — | 45% |
Multilingual GPT-4.5 leads
DeepSeek-V3: 48.5 (#143), GPT-4.5: 52.5 (#83)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| LMArena Non-English | 1358 | 1413 |
| LMArena Chinese | 1391 | 1421 |
| LMArena French | 1385 | 1418 |
| LMArena German | 1374 | 1457 |
| LMArena Japanese | 1333 | 1416 |
| LMArena Korean | 1319 | 1392 |
| LMArena Russian | 1373 | 1419 |
| LMArena Spanish | 1358 | — |
Instruction Following Too close to call
DeepSeek-V3: 72.8 (#130), GPT-4.5: 72.6 (#134)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| LiveBench Instruction Following | 81.5% | 72.3% |
| LMArena Instruction Following | 1345 | 1404 |
| IFEval | 83.2% | — |
Long Context GPT-4.5 leads
DeepSeek-V3: 34.0 (#253), GPT-4.5: 40.4 (#155)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| Fiction.LiveBench | 50% | 63.9% |
| LMArena Longer Query | 1352 | 1406 |
Writing & Preference Too close to call
DeepSeek-V3: 57.4 (#130), GPT-4.5: 56.9 (#134)
| Benchmark | DeepSeek-V3 | GPT-4.5 |
|---|---|---|
| LMArena Text | 1375 | 1417 |
| LMArena Creative Writing | 1364 | 1394 |
| Short-Story Creative Writing | 77% | 75.6% |
| EQ-Bench Creative Writing | 1472 | 1258 |
| LMArena Multi-Turn | 1389 | 1444 |
| LiveBench Language | 49.1% | 61.5% |
| WildBench | 83% | — |
Frequently asked questions
Is DeepSeek-V3 better than GPT-4.5?
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 37.2 on the Noometry Index.
Is DeepSeek-V3 or GPT-4.5 better for coding?
They score almost the same on coding (42.3 vs 42.2); test both on your own repository before choosing.
How many benchmarks do DeepSeek-V3 and GPT-4.5 share?
35 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and GPT-4.5 has 42.