Model comparison
DeepSeek-V3 vs GPT-4o mini
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 25.5 on the Noometry Index. GPT-4o mini costs 1.5× less per token, which makes it the better buy when DeepSeek-V3's lead doesn't matter for your workload.
Last verified . 48 shared benchmarks.
DeepSeekSummary
- They share 48 benchmarks with published results for both. DeepSeek-V3 scores higher in 7 categories and GPT-4o mini in 1 category; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek-V3 leads 32.1 to 10.4.
- The biggest single-benchmark swing is Aider Polyglot: 55.1% for DeepSeek-V3 and 3.6% for GPT-4o mini.
- GPT-4o mini is cheaper at $0.15 / $0.60 per million input/output tokens, against $0.24 / $0.90 for DeepSeek-V3.
- 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 mini | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 39.5 | 25.5 |
| Released | 2024-12-26 | 2024-07-18 |
| Weights | Open | Proprietary |
| Context window | 164K | 128K |
| Max output | 164K | 16K |
| Input $ / M tokens | $0.24 | $0.15 |
| Output $ / M tokens | $0.90 | $0.60 |
| Results tracked | 60 | 60 |
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 mini: 22.0 (#335)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| Aider Polyglot | 55.1% | 3.6% |
| WeirdML | 36.1% | 11.8% |
| BigCodeBench Instruct | 50% | 46.1% |
| LiveBench Coding | 70.9% | 43.1% |
| LMArena Coding | 1368 | 1290 |
| BigCodeBench Complete | 62.2% | 57.4% |
| HumanEval+ | 86.6% | 83.5% |
| MBPP+ | 73% | 72.2% |
| SciCode | 35.8% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, GPT-4o mini: 27.5 (#101)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| BALROG | — | 17.4% |
| METR Time Horizons | 49.6% | — |
Reasoning DeepSeek-V3 leads
DeepSeek-V3: 20.5 (#236), GPT-4o mini: 8.7 (#347)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| SimpleBench | 27.2% | 10.7% |
| Kagi LLM Benchmark | 52.3% | 28.8% |
| LiveBench Reasoning | 65.8% | 32.8% |
| LMArena Hard Prompts | 1365 | 1267 |
| DTBench | 64.8% | 54.4% |
| LiveBench Data Analysis | 60.9% | 50% |
| LMCA | 15.5% | 10.4% |
| Epoch Capabilities Index | 135.94 | 126.56 |
| LiveBench | 66.9% | 41.3% |
| PIQA | 84.7% | 88.7% |
| ARC-AGI-2 | — | 0% |
| CritPt | 0% | — |
| Chess Puzzles | — | 0% |
| Mystery Game Puzzles | — | 12% |
| BIG-Bench Hard | 87.5% | — |
| ForecastBench | 59.1 | — |
| HellaSwag | 88.9% | — |
| WinoGrande | 85.2% | — |
Math DeepSeek-V3 leads
DeepSeek-V3: 32.1 (#219), GPT-4o mini: 10.4 (#314)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 6.9% |
| Omni-MATH | 40.3% | 28% |
| LiveBench Math | 73.5% | 36.3% |
| LMArena Math | 1373 | 1267 |
| MATH Level 5 | 75.5% | 52.6% |
| FrontierMath (Tiers 1-3) | — | 0.7% |
| FrontierMath (Feb 2025 set) | 1.7% | — |
| GSM8K | — | 91.3% |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#155), GPT-4o mini: 17.7 (#284)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| GPQA Diamond | 67.6% | 37.7% |
| MMLU-Pro | 72.3% | 60.3% |
| Confabulations | 26.1% | 37.2% |
| GPQA (HELM) | 53.8% | 36.8% |
| LMArena Expert | 1351 | 1235 |
| MMLU | 87.2% | 81.8% |
| SimpleQA Verified | — | 8.3% |
| Vectara Hallucination Rate | 6.1% | — |
| ARC (AI2) Challenge | 95.3% | — |
| BoolQ | — | 88.7% |
| TriviaQA | 82.9% | — |
Multimodal Not comparable
DeepSeek-V3: —, GPT-4o mini: 25.9 (#122)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| LMArena Vision | — | 1066 |
| Video-MME | — | 64.8% |
| GeoBench | — | 64% |
| VPCT | — | 34% |
Multilingual DeepSeek-V3 leads
DeepSeek-V3: 48.5 (#143), GPT-4o mini: 42.0 (#199)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| LMArena Non-English | 1358 | 1266 |
| LMArena Chinese | 1391 | 1265 |
| LMArena French | 1385 | 1297 |
| LMArena German | 1374 | 1272 |
| LMArena Japanese | 1333 | 1216 |
| LMArena Korean | 1319 | 1195 |
| LMArena Russian | 1373 | 1275 |
| LMArena Spanish | 1358 | 1276 |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 72.8 (#130), GPT-4o mini: 61.9 (#239)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| LiveBench Instruction Following | 81.5% | 56.8% |
| IFEval | 83.2% | 78.2% |
| LMArena Instruction Following | 1345 | 1258 |
Long Context GPT-4o mini leads
DeepSeek-V3: 34.0 (#253), GPT-4o mini: 39.1 (#186)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| LMArena Longer Query | 1352 | 1289 |
| Fiction.LiveBench | 50% | — |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.4 (#130), GPT-4o mini: 39.5 (#248)
| Benchmark | DeepSeek-V3 | GPT-4o mini |
|---|---|---|
| LMArena Text | 1375 | 1286 |
| LMArena Creative Writing | 1364 | 1268 |
| Short-Story Creative Writing | 77% | 67.2% |
| EQ-Bench Creative Writing | 1472 | 873 |
| WildBench | 83% | 79.1% |
| LMArena Multi-Turn | 1389 | 1285 |
| LiveBench Language | 49.1% | 28.6% |
Frequently asked questions
Is DeepSeek-V3 better than GPT-4o mini?
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 25.5 on the Noometry Index. GPT-4o mini costs 1.5× less per token, which makes it the better buy when DeepSeek-V3's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3 or GPT-4o mini?
GPT-4o mini is cheaper. It lists at $0.15 per million input tokens and $0.60 per million output tokens; DeepSeek-V3 lists at $0.24 and $0.90.
Is DeepSeek-V3 or GPT-4o mini better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 42.3 versus 22.0 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 mini share?
48 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and GPT-4o mini has 60.