Model comparison
Deepseek Coder v2 vs Llama 3-70B
Deepseek Coder v2 is the stronger model overall, scoring 35.9 to 28.8 on the Noometry Index.
Last verified . 22 shared benchmarks.
DeepSeek
MetaSummary
- They share 22 benchmarks with published results for both. Deepseek Coder v2 scores higher in 6 categories and Llama 3-70B in 2 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where Deepseek Coder v2 leads 34.9 to 12.8.
- The biggest single-benchmark swing is BigCodeBench Complete: 59.7% for Deepseek Coder v2 and 54.5% for Llama 3-70B.
Side by side
| Deepseek Coder v2 | Llama 3-70B | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 35.9 | 28.8 |
| Released | 2024-06-17 | 2024-04-18 |
| Weights | Open | Open |
| Context window | — | — |
| Max output | — | — |
| Input $ / M tokens | — | — |
| Output $ / M tokens | — | — |
| Results tracked | 24 | 31 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Deepseek Coder v2 leads
Deepseek Coder v2: 38.1 (#183), Llama 3-70B: 35.8 (#218)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| BigCodeBench Instruct | 48.2% | 43.6% |
| LMArena Coding | 1251 | 1206 |
| BigCodeBench Complete | 59.7% | 54.5% |
| HumanEval+ | 82.3% | 72% |
| MBPP+ | 75.1% | 69% |
Agentic & Tool Use Not comparable
Deepseek Coder v2: —, Llama 3-70B: 21.1 (#139)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| Cybench | — | 5% |
Reasoning Deepseek Coder v2 leads
Deepseek Coder v2: 23.6 (#176), Llama 3-70B: 18.0 (#288)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| LMArena Hard Prompts | 1207 | 1195 |
| WinoGrande | 83.7% | 83.5% |
| Kagi LLM Benchmark | — | 35.1% |
| DTBench | — | 54.2% |
| Epoch Capabilities Index | — | 122.93 |
| ForecastBench | — | 57.1 |
Math Deepseek Coder v2 leads
Deepseek Coder v2: 34.9 (#190), Llama 3-70B: 12.8 (#305)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| LMArena Math | 1241 | 1218 |
| OTIS Mock AIME 2024-2025 | — | 4.3% |
| MATH Level 5 | — | 22.6% |
| GSM8K | 94.5% | — |
Knowledge Deepseek Coder v2 leads
Deepseek Coder v2: 32.3 (#212), Llama 3-70B: 20.8 (#277)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| LMArena Expert | 1181 | 1149 |
| GPQA Diamond | — | 40.6% |
| ARC (AI2) Challenge | 64.3% | — |
| MMLU | — | 79.3% |
Multilingual Deepseek Coder v2 leads
Deepseek Coder v2: 36.3 (#240), Llama 3-70B: 33.6 (#251)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| LMArena Non-English | 1182 | 1142 |
| LMArena Chinese | 1201 | 1114 |
| LMArena French | 1185 | 1232 |
| LMArena German | 1164 | 1169 |
| LMArena Japanese | 1126 | 1017 |
| LMArena Korean | 1104 | 1017 |
| LMArena Russian | 1188 | 1159 |
| LMArena Spanish | 1153 | 1241 |
Instruction Following Too close to call
Deepseek Coder v2: 61.7 (#242), Llama 3-70B: 62.5 (#238)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| LMArena Instruction Following | 1180 | 1194 |
Long Context Deepseek Coder v2 leads
Deepseek Coder v2: 37.0 (#224), Llama 3-70B: 35.6 (#240)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| LMArena Longer Query | 1219 | 1174 |
Writing & Preference Llama 3-70B leads
Deepseek Coder v2: 38.2 (#253), Llama 3-70B: 42.8 (#231)
| Benchmark | Deepseek Coder v2 | Llama 3-70B |
|---|---|---|
| LMArena Text | 1191 | 1221 |
| LMArena Creative Writing | 1120 | 1210 |
| LMArena Multi-Turn | 1177 | 1223 |
Frequently asked questions
Is Deepseek Coder v2 better than Llama 3-70B?
Deepseek Coder v2 is the stronger model overall, scoring 35.9 to 28.8 on the Noometry Index.
Is Deepseek Coder v2 or Llama 3-70B better for coding?
Deepseek Coder v2 scores higher on coding benchmarks: 38.1 versus 35.8 in the Noometry coding category.
How many benchmarks do Deepseek Coder v2 and Llama 3-70B share?
22 benchmarks have published results for both models. Deepseek Coder v2 has 24 scored results on Noometry and Llama 3-70B has 31.