Model comparison
Llama 3.1-70B vs Llama 3.1-8B
Llama 3.1-70B is the stronger model overall, scoring 29.6 to 23.0 on the Noometry Index. Llama 3.1-8B costs 7.0× less per token, which makes it the better buy when Llama 3.1-70B's lead doesn't matter for your workload.
Last verified . 34 shared benchmarks.
MetaSummary
- They share 34 benchmarks with published results for both. Llama 3.1-70B scores higher in 9 categories and Llama 3.1-8B in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where Llama 3.1-70B leads 24.2 to 8.0.
- The biggest single-benchmark swing is MMLU-Pro: 65.3% for Llama 3.1-70B and 40.6% for Llama 3.1-8B.
- Llama 3.1-8B is cheaper at $0.05 / $0.08 per million input/output tokens, against $0.40 / $0.40 for Llama 3.1-70B.
Side by side
| Llama 3.1-70B | Llama 3.1-8B | |
|---|---|---|
| Provider | Meta | Meta |
| Noometry Index | 29.6 | 23.0 |
| Released | 2024-07-23 | 2024-07-23 |
| Weights | Open | Open |
| Context window | 128K | 128K |
| Max output | 4K | 4K |
| Input $ / M tokens | $0.40 | $0.05 |
| Output $ / M tokens | $0.40 | $0.08 |
| Results tracked | 35 | 43 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Llama 3.1-70B leads
Llama 3.1-70B: 30.3 (#296), Llama 3.1-8B: 20.2 (#340)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| WeirdML | 9% | 1.7% |
| BigCodeBench Instruct | 46.1% | 32.8% |
| LMArena Coding | 1260 | 1195 |
| BigCodeBench Complete | 54.8% | 40.5% |
| SciCode | — | 13.2% |
| HumanEval+ | — | 62.8% |
| MBPP+ | — | 55.6% |
Agentic & Tool Use Llama 3.1-70B leads
Llama 3.1-70B: 25.1 (#112), Llama 3.1-8B: 22.5 (#131)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| BALROG | 27.9% | 15.1% |
| Berkeley Function Calling Leaderboard | — | 25.8% |
| TheAgentCompany | 6.9% | — |
Reasoning Llama 3.1-70B leads
Llama 3.1-70B: 21.6 (#220), Llama 3.1-8B: 14.9 (#321)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| LMArena Hard Prompts | 1241 | 1175 |
| DTBench | 60% | 50.9% |
| LMCA | 14.8% | 5.4% |
| Epoch Capabilities Index | 125.92 | 116.57 |
| CritPt | — | 0% |
| Chess Puzzles | — | 0% |
| PIQA | — | 81.2% |
Math Llama 3.1-70B leads
Llama 3.1-70B: 13.5 (#304), Llama 3.1-8B: 10.2 (#317)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 3.6% | 1.7% |
| Omni-MATH | 21% | 13.7% |
| LMArena Math | 1252 | 1179 |
| MATH Level 5 | 36.7% | 22.9% |
| GSM8K | — | 82.4% |
Knowledge Llama 3.1-70B leads
Llama 3.1-70B: 24.2 (#269), Llama 3.1-8B: 8.0 (#307)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| GPQA Diamond | 44.2% | 27% |
| MMLU-Pro | 65.3% | 40.6% |
| GPQA (HELM) | 42.6% | 24.7% |
| LMArena Expert | 1209 | 1144 |
| MMLU | 80.1% | 56.1% |
| BoolQ | — | 82.8% |
Multilingual Llama 3.1-70B leads
Llama 3.1-70B: 38.8 (#225), Llama 3.1-8B: 34.0 (#249)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| LMArena Non-English | 1219 | 1148 |
| LMArena Chinese | 1215 | 1151 |
| LMArena French | 1261 | 1177 |
| LMArena German | 1222 | 1144 |
| LMArena Japanese | 1132 | 1061 |
| LMArena Korean | 1140 | 1053 |
| LMArena Russian | 1234 | 1158 |
| LMArena Spanish | 1253 | 1169 |
Instruction Following Llama 3.1-70B leads
Llama 3.1-70B: 65.3 (#223), Llama 3.1-8B: 58.9 (#258)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| IFEval | 82.1% | 74.3% |
| LMArena Instruction Following | 1231 | 1159 |
Long Context Llama 3.1-70B leads
Llama 3.1-70B: 37.6 (#214), Llama 3.1-8B: 35.8 (#238)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| LMArena Longer Query | 1241 | 1182 |
Writing & Preference Llama 3.1-70B leads
Llama 3.1-70B: 35.4 (#267), Llama 3.1-8B: 29.7 (#290)
| Benchmark | Llama 3.1-70B | Llama 3.1-8B |
|---|---|---|
| LMArena Text | 1261 | 1187 |
| LMArena Creative Writing | 1232 | 1154 |
| EQ-Bench Creative Writing | 784 | 713 |
| WildBench | 75.8% | 68.7% |
| LMArena Multi-Turn | 1256 | 1172 |
Frequently asked questions
Is Llama 3.1-70B better than Llama 3.1-8B?
Llama 3.1-70B is the stronger model overall, scoring 29.6 to 23.0 on the Noometry Index. Llama 3.1-8B costs 7.0× less per token, which makes it the better buy when Llama 3.1-70B's lead doesn't matter for your workload.
Which is cheaper, Llama 3.1-70B or Llama 3.1-8B?
Llama 3.1-8B is cheaper. It lists at $0.05 per million input tokens and $0.08 per million output tokens; Llama 3.1-70B lists at $0.40 and $0.40.
Is Llama 3.1-70B or Llama 3.1-8B better for coding?
Llama 3.1-70B scores higher on coding benchmarks: 30.3 versus 20.2 in the Noometry coding category.
Which has the bigger context window?
Both accept 128K tokens.
How many benchmarks do Llama 3.1-70B and Llama 3.1-8B share?
34 benchmarks have published results for both models. Llama 3.1-70B has 35 scored results on Noometry and Llama 3.1-8B has 43.