Model comparison
Llama 2-7B vs Llama 3-70B
Llama 2-7B and Llama 3-70B score almost the same on the Noometry Index (29.1 vs 28.8), so choose on price, context window or the category you care about most.
Last verified . 18 shared benchmarks.
MetaSummary
- They share 18 benchmarks with published results for both. Llama 2-7B scores higher in 2 categories and Llama 3-70B in 6 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where Llama 2-7B leads 30.7 to 12.8.
Side by side
| Llama 2-7B | Llama 3-70B | |
|---|---|---|
| Provider | Meta | Meta |
| Noometry Index | 29.1 | 28.8 |
| Released | 2023-07-18 | 2024-04-18 |
| Weights | Open | Open |
| Context window | — | — |
| Max output | — | — |
| Input $ / M tokens | — | — |
| Output $ / M tokens | — | — |
| Results tracked | 29 | 31 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Llama 3-70B leads
Llama 2-7B: 29.2 (#307), Llama 3-70B: 35.8 (#218)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| LMArena Coding | 1002 | 1206 |
| BigCodeBench Instruct | — | 43.6% |
| BigCodeBench Complete | — | 54.5% |
| HumanEval+ | — | 72% |
| MBPP+ | — | 69% |
Agentic & Tool Use Not comparable
Llama 2-7B: —, Llama 3-70B: 21.1 (#139)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| Cybench | — | 5% |
Reasoning Llama 3-70B leads
Llama 2-7B: 15.7 (#312), Llama 3-70B: 18.0 (#288)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| LMArena Hard Prompts | 1009 | 1195 |
| Epoch Capabilities Index | 99.06 | 122.93 |
| WinoGrande | 69.2% | 83.5% |
| Kagi LLM Benchmark | — | 35.1% |
| Chess Puzzles | 0% | — |
| DTBench | — | 54.2% |
| BIG-Bench Hard | 39.2% | — |
| ForecastBench | — | 57.1 |
| HellaSwag | 77.2% | — |
| LAMBADA | 73.3% | — |
| PIQA | 78.8% | — |
Math Llama 2-7B leads
Llama 2-7B: 30.7 (#233), Llama 3-70B: 12.8 (#305)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| LMArena Math | 1042 | 1218 |
| OTIS Mock AIME 2024-2025 | — | 4.3% |
| MATH Level 5 | — | 22.6% |
| GSM8K | 16.7% | — |
Knowledge Llama 2-7B leads
Llama 2-7B: 28.2 (#248), Llama 3-70B: 20.8 (#277)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| LMArena Expert | 1036 | 1149 |
| MMLU | 45.8% | 79.3% |
| GPQA Diamond | — | 40.6% |
| ARC (AI2) Challenge | 45.9% | — |
| BoolQ | 77.9% | — |
| OpenBookQA | 58.6% | — |
| TriviaQA | 73.7% | — |
Multimodal Not comparable
Llama 2-7B: —, Llama 3-70B: —
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| ScienceQA | 43.1% | — |
Multilingual Llama 3-70B leads
Llama 2-7B: 23.8 (#293), Llama 3-70B: 33.6 (#251)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| LMArena Non-English | 973 | 1142 |
| LMArena Chinese | 973 | 1114 |
| LMArena French | 970 | 1232 |
| LMArena German | 978 | 1169 |
| LMArena Russian | 995 | 1159 |
| LMArena Spanish | 1007 | 1241 |
| LMArena Japanese | — | 1017 |
| LMArena Korean | — | 1017 |
Instruction Following Llama 3-70B leads
Llama 2-7B: 50.8 (#298), Llama 3-70B: 62.5 (#238)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| LMArena Instruction Following | 1006 | 1194 |
Long Context Llama 3-70B leads
Llama 2-7B: 30.4 (#287), Llama 3-70B: 35.6 (#240)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| LMArena Longer Query | 999 | 1174 |
Writing & Preference Llama 3-70B leads
Llama 2-7B: 28.0 (#298), Llama 3-70B: 42.8 (#231)
| Benchmark | Llama 2-7B | Llama 3-70B |
|---|---|---|
| LMArena Text | 1053 | 1221 |
| LMArena Creative Writing | 1033 | 1210 |
| LMArena Multi-Turn | 1029 | 1223 |
Frequently asked questions
Is Llama 2-7B better than Llama 3-70B?
Llama 2-7B and Llama 3-70B score almost the same on the Noometry Index (29.1 vs 28.8), so choose on price, context window or the category you care about most.
Is Llama 2-7B or Llama 3-70B better for coding?
Llama 3-70B scores higher on coding benchmarks: 35.8 versus 29.2 in the Noometry coding category.
How many benchmarks do Llama 2-7B and Llama 3-70B share?
18 benchmarks have published results for both models. Llama 2-7B has 29 scored results on Noometry and Llama 3-70B has 31.