Mistral: Mixtral 8x22B Instruct

Text input Text output
Author's Description

Mistral's official instruct fine-tuned version of [Mixtral 8x22B](/models/mistralai/mixtral-8x22b). It uses 39B active parameters out of 141B, offering unparalleled cost efficiency for its size. Its strengths include: - strong math, coding, and reasoning - large context length (64k) - fluency in English, French, Italian, German, and Spanish See benchmarks on the launch announcement [here](https://mistral.ai/news/mixtral-8x22b/). #moe

Key Specifications
Cost
$$$
Context
65K
Parameters
22B
Released
Apr 16, 2024
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Tool Choice Response Format Top P Temperature Top Logprobs Tools Logit Bias Structured Outputs Logprobs Stop Max Tokens Frequency Penalty Presence Penalty
Features

This model supports the following features:

Tools Response Format Structured Outputs
Performance Summary

Mistral's Mixtral 8x22B Instruct model, released on April 16, 2024, demonstrates a strong overall performance profile, particularly excelling in speed and reliability. It consistently performs among the fastest models, ranking in the 80th percentile across eight benchmarks. The model also offers competitive pricing, positioned at the 48th percentile. Notably, its reliability is exceptional, achieving a 100% success rate across all benchmarks, indicating minimal technical failures. In terms of specific benchmark performance, Mixtral 8x22B Instruct shows perfect accuracy in Hallucinations (Baseline), making it the most accurate model at its price point and speed. It exhibits strong capabilities in General Knowledge (95.5% accuracy), Ethics (95.0% accuracy), and Mathematics (86.0% accuracy). Its stated strengths in math, coding, and reasoning are largely supported, with Coding achieving 83.0% accuracy. However, the model shows weaknesses in Email Classification (80.0% accuracy) and particularly in Instruction Following (50.0% accuracy) and Reasoning (54.0% accuracy), where its performance falls below the 50th percentile. Its large context length of 65536 and fluency in multiple languages further enhance its utility.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.9
Completion $0.9

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Fireworks
Fireworks | mistralai/mixtral-8x22b-instruct 65K $0.9 / 1M tokens $0.9 / 1M tokens
Mistral
Mistral | mistralai/mixtral-8x22b-instruct 65K $2 / 1M tokens $6 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Strategy Free Executions Accuracy Cost Duration
Other Models by mistralai