Model comparison
Llama 2-70B vs Llama 3-8B
Llama 3-8B is the stronger model overall, scoring 25.5 to 24.4 on the Noometry Index.
Last verified . 28 shared benchmarks.
MetaSummary
- They share 28 benchmarks with published results for both. Llama 2-70B scores higher in 2 categories and Llama 3-8B in 6 categories; 4 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where Llama 3-8B leads 37.5 to 32.3.
Side by side
| Llama 2-70B | Llama 3-8B | |
|---|---|---|
| Provider | Meta | Meta |
| Noometry Index | 24.4 | 25.5 |
| Released | 2023-07-18 | 2024-04-18 |
| Weights | Open | Open |
| Context window | — | — |
| Max output | — | — |
| Input $ / M tokens | — | — |
| Output $ / M tokens | — | — |
| Results tracked | 35 | 34 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
Llama 2-70B: 31.4 (#286), Llama 3-8B: 31.0 (#289)
| Benchmark | Llama 2-70B | Llama 3-8B |
|---|---|---|
| LMArena Coding | 1079 | 1152 |
| BigCodeBench Instruct | — | 31.9% |
| BigCodeBench Complete | — | 36.9% |
| HumanEval+ | — | 56.7% |
| MBPP+ | — | 54.8% |
Reasoning Too close to call
Llama 2-70B: 14.4 (#325), Llama 3-8B: 14.3 (#326)
| Benchmark | Llama 2-70B | Llama 3-8B |
|---|---|---|
| LMArena Hard Prompts | 1073 | 1133 |
| DTBench | 41.6% | 43.9% |
| Epoch Capabilities Index | 113.79 | 116.45 |
| ForecastBench | 51.4 | 58.6 |
| WinoGrande | 80.2% | 75.7% |
| Chess Puzzles | — | 0% |
| Adversarial NLI | — | 57.3% |
| BIG-Bench Hard | 64.9% | — |
| CommonsenseQA 2.0 | 50% | — |
| HellaSwag | 85.3% | — |
| LAMBADA | 78.9% | — |
| PIQA | 82.8% | — |
Math Too close to call
Llama 2-70B: 8.1 (#326), Llama 3-8B: 8.8 (#323)
| Benchmark | Llama 2-70B | Llama 3-8B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 0% | 1.9% |
| LMArena Math | 1091 | 1151 |
| MATH Level 5 | 3.3% | 6.1% |
| GSM8K | 69.6% | — |
Knowledge Too close to call
Llama 2-70B: 7.4 (#310), Llama 3-8B: 7.8 (#308)
| Benchmark | Llama 2-70B | Llama 3-8B |
|---|---|---|
| GPQA Diamond | 26.3% | 26.1% |
| LMArena Expert | 1039 | 1113 |
| ARC (AI2) Challenge | 78.3% | 82.8% |
| MMLU | 69.9% | 68.8% |
| OpenBookQA | 60.2% | 82.6% |
| TriviaQA | 87.6% | 67.7% |
| BoolQ | 88.6% | — |
Multilingual Llama 3-8B leads
Llama 2-70B: 27.7 (#274), Llama 3-8B: 30.8 (#261)
| Benchmark | Llama 2-70B | Llama 3-8B |
|---|---|---|
| LMArena Non-English | 1045 | 1098 |
| LMArena Chinese | 995 | 1076 |
| LMArena French | 1090 | 1159 |
| LMArena German | 1041 | 1104 |
| LMArena Japanese | 927 | 967 |
| LMArena Korean | 964 | 1004 |
| LMArena Russian | 1083 | 1109 |
| LMArena Spanish | 1143 | 1173 |
Instruction Following Llama 3-8B leads
Llama 2-70B: 54.9 (#278), Llama 3-8B: 58.4 (#260)
| Benchmark | Llama 2-70B | Llama 3-8B |
|---|---|---|
| LMArena Instruction Following | 1071 | 1127 |
Long Context Llama 3-8B leads
Llama 2-70B: 32.3 (#270), Llama 3-8B: 34.2 (#251)
| Benchmark | Llama 2-70B | Llama 3-8B |
|---|---|---|
| LMArena Longer Query | 1062 | 1128 |
Writing & Preference Llama 3-8B leads
Llama 2-70B: 32.3 (#279), Llama 3-8B: 37.5 (#256)
| Benchmark | Llama 2-70B | Llama 3-8B |
|---|---|---|
| LMArena Text | 1115 | 1166 |
| LMArena Creative Writing | 1075 | 1150 |
| LMArena Multi-Turn | 1088 | 1152 |
Frequently asked questions
Is Llama 2-70B better than Llama 3-8B?
Llama 3-8B is the stronger model overall, scoring 25.5 to 24.4 on the Noometry Index.
Is Llama 2-70B or Llama 3-8B better for coding?
They score almost the same on coding (31.4 vs 31.0); test both on your own repository before choosing.
How many benchmarks do Llama 2-70B and Llama 3-8B share?
28 benchmarks have published results for both models. Llama 2-70B has 35 scored results on Noometry and Llama 3-8B has 34.