Mixtral 8x7B
Mixtral 8x7B
Section titled “Mixtral 8x7B”Provider: Groq (Mistral) Context Window: 32,000 tokens Quality: Medium Price: $0.27 per 1M input tokens
Overview
Section titled “Overview”Mixtral 8x7B is Mistral’s mixture-of-experts model, offering good performance at low cost. It uses a sparse architecture that activates only relevant parameters for each token.
Benchmarks
Section titled “Benchmarks”| Benchmark | Score | Percentile |
|---|---|---|
| MMLU | 77.3% | Top 20% |
| HumanEval | 74.4% | Top 20% |
| MATH | 52.2% | Top 30% |
| GPQA | 31.6% | Top 35% |
| HellaSwag | 90.8% | Top 20% |
| ARC-Challenge | 90.2% | Top 20% |
Source: Mistral AI technical reports
Best For
Section titled “Best For”- Balanced quality and cost
- Multi-lingual tasks
- General-purpose workloads
- Cost-conscious teams
Pricing
Section titled “Pricing”| Tier | Input | Output |
|---|---|---|
| BYOK | $0.27/1M | $0.27/1M |
| Marketplace | $0.2781/1M | $0.2781/1M |
Routing Recommendation
Section titled “Routing Recommendation”- Accurate mode: Occasionally selected for multi-lingual tasks
- Balanced mode: Good default for cost-aware users
- Cheap mode: Primary choice for Mistral users
- Eco mode: Frequently selected
curl http://127.0.0.1:8787/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{"model": "fastlane-auto", "messages": [{"role": "user", "content": "Translate this to French"}]}'