O

LangMart: Mistral: Mixtral 8x7B Instruct

Openrouter
33K
Context
$0.5400
Input /1M
$0.5400
Output /1M
N/A
Max Output

LangMart: Mistral: Mixtral 8x7B Instruct

Model Overview

Property Value
Model ID openrouter/mistralai/mixtral-8x7b-instruct
Name Mistral: Mixtral 8x7B Instruct
Provider mistralai
Released 2023-12-10

Description

Mixtral 8x7B Instruct is a pretrained generative Sparse Mixture of Experts, by Mistral AI, for chat and instruction use. Incorporates 8 experts (feed-forward networks) for a total of 47 billion parameters.

Instruct model fine-tuned by Mistral. #moe

Description

LangMart: Mistral: Mixtral 8x7B Instruct is a language model provided by mistralai. This model offers advanced capabilities for natural language processing tasks.

Provider

mistralai

Specifications

Spec Value
Context Window 32,768 tokens
Modalities text->text
Input Modalities text
Output Modalities text

Pricing

Type Price
Input $0.54 per 1M tokens
Output $0.54 per 1M tokens

Capabilities

  • Frequency penalty
  • Logit bias
  • Max tokens
  • Min p
  • Presence penalty
  • Repetition penalty
  • Response format
  • Seed
  • Stop
  • Temperature
  • Tool choice
  • Tools
  • Top k
  • Top p

Detailed Analysis

Mixtral 8x7B Instruct is Mistral's groundbreaking Sparse Mixture of Experts (SMoE) model featuring 8 expert networks of 7B parameters each (47B total parameters) but only activating 13B parameters per token through intelligent routing. This architectural innovation delivers near-50B model quality at 13B computational cost - approximately 6x faster inference than dense 70B models while outperforming Llama 2 70B on most benchmarks. The router network dynamically selects the 2 most relevant experts per token, enabling specialization (one expert for code, another for math, etc.) while maintaining efficiency. Mixtral 8x7B excels at diverse tasks: complex code generation, mathematical reasoning, multilingual understanding (supporting 32K context across languages), and general reasoning. The model supports function calling and JSON mode for agentic applications. Released under Apache 2.0, it revolutionized open-source AI by proving MoE architectures could match proprietary models. Ideal for applications requiring near-frontier performance at fraction of computational cost, self-hosting scenarios, and research into MoE architectures.