Model comparison
DeepSeek-V3 vs Llama 3.1-70B
DeepSeek-V3 is the stronger model overall, scoring 39.2 to 29.4 on the Noometry Index.
Last verified . 33 shared benchmarks.
DeepSeek
MetaSummary
- They share 33 benchmarks with published results for both. DeepSeek-V3 scores higher in 6 categories and Llama 3.1-70B in 2 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where DeepSeek-V3 leads 57.6 to 35.6.
- The biggest single-benchmark swing is MATH Level 5: 75.5% for DeepSeek-V3 and 36.7% for Llama 3.1-70B.
- Both cost about the same: $0.24 input and $0.90 output per million tokens.
- DeepSeek-V3 accepts more context: 164K tokens versus 128K.
Side by side
| DeepSeek-V3 | Llama 3.1-70B | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 39.2 | 29.4 |
| Released | 2024-12-26 | 2024-07-23 |
| Weights | Open | Open |
| Context window | 164K | 128K |
| Max output | 164K | 4K |
| Input $ / M tokens | $0.24 | $0.40 |
| Output $ / M tokens | $0.90 | $0.40 |
| Results tracked | 60 | 35 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3 leads
DeepSeek-V3: 41.9 (#99), Llama 3.1-70B: 29.8 (#292)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| WeirdML | 36.1% | 9% |
| BigCodeBench Instruct | 50% | 46.1% |
| LMArena Coding | 1368 | 1260 |
| BigCodeBench Complete | 62.2% | 54.8% |
| Aider Polyglot | 55.1% | — |
| SciCode | 35.8% | — |
| LiveBench Coding | 70.9% | — |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, Llama 3.1-70B: 24.9 (#104)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| TheAgentCompany | — | 6.9% |
| BALROG | — | 27.9% |
| METR Time Horizons | 49.6% | — |
Reasoning Too close to call
DeepSeek-V3: 20.0 (#226), Llama 3.1-70B: 20.8 (#215)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| LMArena Hard Prompts | 1365 | 1241 |
| DTBench | 64.8% | 60% |
| LMCA | 15.5% | 14.8% |
| Epoch Capabilities Index | 135.95 | 125.91 |
| SimpleBench | 27.2% | — |
| Kagi LLM Benchmark | 52.3% | — |
| CritPt | 0% | — |
| LiveBench Reasoning | 65.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| BIG-Bench Hard | 87.5% | — |
| ForecastBench | 59.1 | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math DeepSeek-V3 leads
DeepSeek-V3: 30.4 (#208), Llama 3.1-70B: 12.7 (#298)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 3.6% |
| Omni-MATH | 40.3% | 21% |
| LMArena Math | 1373 | 1252 |
| MATH Level 5 | 75.5% | 36.7% |
| LiveBench Math | 73.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#147), Llama 3.1-70B: 24.1 (#264)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| GPQA Diamond | 67.6% | 44.2% |
| MMLU-Pro | 72.3% | 65.3% |
| GPQA (HELM) | 53.8% | 42.6% |
| LMArena Expert | 1352 | 1209 |
| MMLU | 87.2% | 80.1% |
| Confabulations | 26.1% | — |
| Vectara Hallucination Rate | 6.1% | — |
| ARC (AI2) Challenge | 95.3% | — |
| TriviaQA | 82.9% | — |
Multilingual DeepSeek-V3 leads
DeepSeek-V3: 48.8 (#138), Llama 3.1-70B: 38.9 (#220)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| LMArena Non-English | 1358 | 1219 |
| LMArena Chinese | 1390 | 1215 |
| LMArena French | 1386 | 1260 |
| LMArena German | 1374 | 1222 |
| LMArena Japanese | 1333 | 1132 |
| LMArena Korean | 1321 | 1140 |
| LMArena Russian | 1372 | 1234 |
| LMArena Spanish | 1358 | 1253 |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 73.0 (#125), Llama 3.1-70B: 65.5 (#218)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| IFEval | 83.2% | 82.1% |
| LMArena Instruction Following | 1345 | 1231 |
| LiveBench Instruction Following | 81.5% | — |
Long Context Llama 3.1-70B leads
DeepSeek-V3: 34.0 (#248), Llama 3.1-70B: 37.6 (#208)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| LMArena Longer Query | 1353 | 1241 |
| Fiction.LiveBench | 50% | — |
Writing & Preference DeepSeek-V3 leads
DeepSeek-V3: 57.6 (#124), Llama 3.1-70B: 35.6 (#262)
| Benchmark | DeepSeek-V3 | Llama 3.1-70B |
|---|---|---|
| LMArena Text | 1375 | 1261 |
| LMArena Creative Writing | 1364 | 1232 |
| EQ-Bench Creative Writing | 1472 | 784 |
| WildBench | 83% | 75.8% |
| LMArena Multi-Turn | 1389 | 1256 |
| Short-Story Creative Writing | 77% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than Llama 3.1-70B?
DeepSeek-V3 is the stronger model overall, scoring 39.2 to 29.4 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or Llama 3.1-70B?
Llama 3.1-70B is cheaper. It lists at $0.40 per million input tokens and $0.40 per million output tokens; DeepSeek-V3 lists at $0.24 and $0.90.
Is DeepSeek-V3 or Llama 3.1-70B better for coding?
DeepSeek-V3 scores higher on coding benchmarks: 41.9 versus 29.8 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 Llama 3.1-70B share?
33 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and Llama 3.1-70B has 35.