Llama 2 70B vs Llama 3 70B: Benchmark Comparison
Detailed comparison of Llama 2 70B and Llama 3 70B covering benchmarks, pricing, context window, and compliance.
Key Specifications
| Specification | Llama 2 70B | Llama 3 70B |
|---|---|---|
| Vendor | meta | meta |
| Version | 2-70b | 3-70b |
| Release Date | 2023-07-18 | 2024-04-18 |
| Context Window | 4096 tokens | 8192 tokens |
| Input Modalities | text | text |
| Output Modalities | text | text |
| License | Llama 2 Community License | Llama 3 Community License |
| SOC2 | ✗ | ✗ |
| HIPAA | ✗ | ✗ |
| GDPR | ✗ | ✗ |
| ISO 27001 | ✗ | ✗ |
Benchmark Results
| Benchmark | Llama 2 70B | Llama 3 70B | Winner |
|---|---|---|---|
| ARC | 80.3 | 93.4 | Llama 3 70B |
| BBH | 62.5 | 77.6 | Llama 3 70B |
| GPQA | 27.2 | 38.2 | Llama 3 70B |
| GSM8K | 51.8 | 76.2 | Llama 3 70B |
| HUMANEVAL | 50.6 | 73 | Llama 3 70B |
| IFEVAL | 57.1 | 72.2 | Llama 3 70B |
| MATH | 22.4 | 50.1 | Llama 3 70B |
| MMLU | 53 | 79.5 | Llama 3 70B |
| MUSR | 36.1 | 51.2 | Llama 3 70B |
| WINOGRANDE | 73.4 | 79.2 | Llama 3 70B |
Pricing Comparison
| Tier (per Mtok) | Llama 2 70B | Llama 3 70B |
|---|---|---|
| Input | $0.9 | $0.9 |
| Output | $0.9 | $0.9 |
| Cache Read | $0 | $0 |
| Cache Write | $0 | $0 |
Llama 2 70B contra Llama 3 70B
Visão geral do modelo
Llama 2 70B and Llama 3 70B are both notable options in the AI model market. This page compares their benchmarks, pricing, and compliance.
Especificações principais
| Fornecedor | Data de lançamento | Janela de contexto | Licença |
|---|---|---|---|
| Meta / Meta | 2023-07-18 / 2024-04-18 | 4K / 8K | Llama 2 Community License / Llama 3 Community License |
Desempenho em benchmarks
| Benchmark | Llama 2 70B | Llama 3 70B | Vencedor |
|---|---|---|---|
| ARC | 80.3 | 93.4 | B |
| BBH (BIG-Bench Hard) | 62.5 | 77.6 | B |
| GPQA | 27.2 | 38.2 | B |
| GSM8K (Grade School Math 8K) | 51.8 | 76.2 | B |
| HumanEval | 50.6 | 73.0 | B |
| IFEval | 57.1 | 72.2 | B |
| MATH | 22.4 | 50.1 | B |
| MMLU (Massive Multitask Language Understanding) | 53.0 | 79.5 | B |
| MUSR | 36.1 | 51.2 | B |
| WinoGrande | 73.4 | 79.2 | B |
Comparação de preços
| Entrada | Saída | Leitura de cache | Escrita de cache |
|---|---|---|---|
| — / — | — / — | — / — | — / — |
por milhão de tokens — A / B
Pontos fortes & Pontos fracos
Llama 2 70B
- ✅ 可靠的通用模型。
- ⚠️ MMLU 仅 53.0，知识推理偏弱。
- ⚠️ 闭源专有模型，不支持自托管。
- ⚠️ 上下文窗口 4K 偏小。
Llama 3 70B
- ✅ 可靠的通用模型。
- ⚠️ 闭源专有模型，不支持自托管。
- ⚠️ 上下文窗口 8K 偏小。
Opinião do editor
Llama 2 70B and Llama 3 70B each have their strengths. Choose based on workload (code, long context, vision), referencing the tables above.
Perguntas frequentes
Which model is better for coding tasks?
Refer to the HumanEval benchmark table; the model with a higher score is better suited for coding tasks.
Which model is cheaper?
Refer to the pricing comparison table above; the model with lower input/output prices is more cost-effective.
Which has a longer context window?
Refer to the key specifications table; the model with a larger context window is better for long documents.
Referências
Editor's Take
See Editor's Take section.