Mixtral 8x7B vs Mixtral 8x22B: Benchmark Comparison
Detailed comparison of Mixtral 8x7B and Mixtral 8x22B covering benchmarks, pricing, context window, and compliance.
Key Specifications
| Specification | Mixtral 8x7B | Mixtral 8x22B |
|---|---|---|
| Vendor | mistral | mistral |
| Version | 8x7b | 8x22b |
| Release Date | 2023-12-11 | 2024-04-10 |
| Context Window | 32000 tokens | 64000 tokens |
| Input Modalities | text | text |
| Output Modalities | text | text |
| License | Apache 2.0 | Apache 2.0 |
| SOC2 | ✓ | ✗ |
| HIPAA | ✗ | ✗ |
| GDPR | ✓ | ✗ |
| ISO 27001 | ✓ | ✗ |
Benchmark Results
| Benchmark | Mixtral 8x7B | Mixtral 8x22B | Winner |
|---|---|---|---|
| ARC | 91.6 | — | Mixtral 8x7B |
| BBH | 78.4 | 74.5 | Mixtral 8x7B |
| GPQA | 43 | — | Mixtral 8x7B |
| GSM8K | 79.7 | 78.6 | Mixtral 8x7B |
| HUMANEVAL | 79 | 45.2 | Mixtral 8x7B |
| IFEVAL | 72.1 | — | Mixtral 8x7B |
| MATH | 38 | 46 | Mixtral 8x22B |
| MMLU | 77.2 | 77.8 | Mixtral 8x22B |
| MUSR | 53.5 | — | Mixtral 8x7B |
| WINOGRANDE | 82 | — | Mixtral 8x7B |
Pricing Comparison
| Tier (per Mtok) | Mixtral 8x7B | Mixtral 8x22B |
|---|---|---|
| Input | $0.7 | $1.2 |
| Output | $0.7 | $1.2 |
| Cache Read | $0 | $0 |
| Cache Write | $0 | $0 |
Mixtral 8x7B против Mixtral 8x22B
Обзор модели
Mixtral 8x7B and Mixtral 8x22B are both notable options in the AI model market. This page compares their benchmarks, pricing, and compliance.
Ключевые характеристики
| Поставщик | Дата выпуска | Окно контекста | Лицензия |
|---|---|---|---|
| Mistral / Mistral | 2023-12-11 / 2024-04-10 | 32K / 64K | Apache 2.0 / Apache 2.0 |
Производительность
| Бенчмарк | Mixtral 8x7B | Mixtral 8x22B | Победитель |
|---|---|---|---|
| ARC | 91.6 | — | A |
| BBH (BIG-Bench Hard) | 78.4 | 74.5 | A |
| GPQA | 43.0 | — | A |
| GSM8K (Grade School Math 8K) | 79.7 | 78.6 | A |
| HumanEval | 79.0 | 45.2 | A |
| IFEval | 72.1 | — | A |
| MATH | 38.0 | 46.0 | B |
| MMLU (Massive Multitask Language Understanding) | 77.2 | 77.8 | B |
| MUSR | 53.5 | — | A |
| WinoGrande | 82.0 | — | A |
Сравнение цен
| Вход | Выход | Чтение кэша | Запись кэша |
|---|---|---|---|
| — / — | — / — | — / — | — / — |
за миллион токенов — A / B
Сильные стороны & Слабые стороны
Mixtral 8x7B
- ✅ 采用 MoE 混合专家架构。
- ⚠️ 闭源专有模型，不支持自托管。
Mixtral 8x22B
- ✅ 采用 MoE 混合专家架构。
- ⚠️ HumanEval 45.2，代码能力较弱。
- ⚠️ 闭源专有模型，不支持自托管。
Мнение редактора
Mixtral 8x7B and Mixtral 8x22B each have their strengths. Choose based on workload (code, long context, vision), referencing the tables above.
ЧЗВ
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.
Ссылки
Editor's Take
See Editor's Take section.