Model comparison
DeepSeek-V3 vs Llama 3.1-405B
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 30.7 on the Noometry Index.
Last verified . 40 shared benchmarks.
DeepSeek
MetaSummary
- They share 40 benchmarks with published results for both. DeepSeek-V3 scores higher in 7 categories and Llama 3.1-405B in 1 category; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where DeepSeek-V3 leads 57.4 to 38.9.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 37.8% for DeepSeek-V3 and 9.7% for Llama 3.1-405B.
Side by side
| DeepSeek-V3 | Llama 3.1-405B | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 39.5 | 30.7 |
| Released | 2024-12-26 | 2024-07-23 |
| Weights | Open | Open |
| 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 DeepSeek-V3 leads
DeepSeek-V3: 42.3 (#106), Llama 3.1-405B: 33.1 (#262)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| WeirdML | 36.1% | 21.4% |
| LMArena Coding | 1368 | 1291 |
| Aider Polyglot | 55.1% | — |
| SciCode | 35.8% | — |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, Llama 3.1-405B: 21.0 (#140)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| TheAgentCompany | — | 7.4% |
| Cybench | — | 7.5% |
| METR Time Horizons | 49.6% | — |
Reasoning DeepSeek-V3 leads
DeepSeek-V3: 20.5 (#236), Llama 3.1-405B: 16.8 (#300)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| SimpleBench | 27.2% | 23% |
| Kagi LLM Benchmark | 52.3% | 45% |
| LMArena Hard Prompts | 1365 | 1269 |
| DTBench | 64.8% | 61.4% |
| BIG-Bench Hard | 87.5% | 82.9% |
| Epoch Capabilities Index | 135.94 | 128.75 |
| ForecastBench | 59.1 | 59.9 |
| HellaSwag | 88.9% | 89.2% |
| PIQA | 84.7% | 85.9% |
| WinoGrande | 85.2% | 89.2% |
| CritPt | 0% | — |
| LiveBench Reasoning | 65.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| LMCA | 15.5% | — |
| LiveBench | 66.9% | — |
Math DeepSeek-V3 leads
DeepSeek-V3: 32.1 (#219), Llama 3.1-405B: 18.4 (#290)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 9.7% |
| Omni-MATH | 40.3% | 24.9% |
| LMArena Math | 1373 | 1281 |
| MATH Level 5 | 75.5% | 49.8% |
| LiveBench Math | 73.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#155), Llama 3.1-405B: 30.4 (#227)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| GPQA Diamond | 67.6% | 50.9% |
| MMLU-Pro | 72.3% | 72.3% |
| Confabulations | 26.1% | 17.6% |
| GPQA (HELM) | 53.8% | 52.2% |
| LMArena Expert | 1351 | 1243 |
| ARC (AI2) Challenge | 95.3% | 95.3% |
| MMLU | 87.2% | 84.5% |
| TriviaQA | 82.9% | 82.7% |
| Vectara Hallucination Rate | 6.1% | — |
Multilingual DeepSeek-V3 leads
DeepSeek-V3: 48.5 (#143), Llama 3.1-405B: 40.7 (#214)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| LMArena Non-English | 1358 | 1248 |
| LMArena Chinese | 1391 | 1242 |
| LMArena French | 1385 | 1279 |
| LMArena German | 1374 | 1252 |
| LMArena Japanese | 1333 | 1208 |
| LMArena Korean | 1319 | 1184 |
| LMArena Russian | 1373 | 1265 |
| LMArena Spanish | 1358 | 1260 |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 72.8 (#130), Llama 3.1-405B: 65.9 (#214)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| IFEval | 83.2% | 81.1% |
| LMArena Instruction Following | 1345 | 1259 |
| LiveBench Instruction Following | 81.5% | — |
Long Context Llama 3.1-405B leads
DeepSeek-V3: 34.0 (#253), Llama 3.1-405B: 38.4 (#197)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| LMArena Longer Query | 1352 | 1266 |
| Fiction.LiveBench | 50% | — |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.4 (#130), Llama 3.1-405B: 38.9 (#251)
| Benchmark | DeepSeek-V3 | Llama 3.1-405B |
|---|---|---|
| LMArena Text | 1375 | 1284 |
| LMArena Creative Writing | 1364 | 1262 |
| EQ-Bench Creative Writing | 1472 | 870 |
| WildBench | 83% | 78.3% |
| LMArena Multi-Turn | 1389 | 1297 |
| Short-Story Creative Writing | 77% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than Llama 3.1-405B?
DeepSeek-V3 is the stronger model overall, scoring 39.5 to 30.7 on the Noometry Index.
Is DeepSeek-V3 or Llama 3.1-405B better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 42.3 versus 33.1 in the Noometry coding category.
How many benchmarks do DeepSeek-V3 and Llama 3.1-405B share?
40 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and Llama 3.1-405B has 42.