Neura Market
Mixtral-8x22B logo

Mixtral-8x22B

Free

A sparse MoE model setting new standards for performance and efficiency.

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
Founded
2023
Company
Mistral AI
Visit Website GitHub (38 stars)

About Mixtral-8x22B

Mixtral 8x22B is a sparse mixture-of-experts (SMoE) language model developed by Mistral AI, released in April 2024 under the Apache 2.0 open-source license. It has 141B total parameters with only 39B active per token, offering a strong performance-to-cost ratio. The model supports a 64K token context window, is natively fluent in English, French, Italian, German, and Spanish, and excels at mathematics, coding, and reasoning tasks. It also natively supports function calling and constrained output modes, making it suitable for application development and enterprise tech stack modernization. The base model is available for fine-tuning, and the instructed version achieves state-of-the-art results on benchmarks like GSM8K (90.8%) and Math (44.6%).

Key Features

Sparse Mixture-of-Experts (SMoE) architecture with 141B total parameters, 39B active per token
64K token context window for precise information recall from large documents
Native multilingual support: English, French, Italian, German, Spanish
Strong mathematical and coding capabilities (90.8% on GSM8K, 44.6% on Math with instructed version)
Native function calling and constrained output mode
Released under Apache 2.0, the most permissive open-source license
Optimized for reasoning and outperforms other open-weight models
Cost-efficient with best performance-to-cost ratio among open models of similar size

Pros & Cons

Pros
  • Open-source under Apache 2.0 with no usage restrictions
  • Excellent performance relative to active parameter count (39B active of 141B total)
  • Faster inference than dense 70B models due to sparse activation
  • Strong multilingual and math/coding capabilities
  • Large 64K context window for handling long documents
  • Supports function calling and constrained output for production use
Cons
  • Requires significant computational resources for inference due to large total parameter count (141B)
  • Not a full product with chat interface; requires self-hosting or integration
  • May not surpass the absolute performance of the largest closed-source frontier models on all tasks

Best For

Application development and tech stack modernization via function callingMulti-step reasoning and complex problem solvingMultilingual NLP tasks across five languagesCode generation, code completion, and debuggingFine-tuning for domain-specific tasksLarge document analysis and summarization with extended contextEducational and research applications requiring open-source models

FAQ

What license is Mixtral 8x22B released under?
Mixtral 8x22B is released under the Apache 2.0 license, a permissive open-source license that allows unrestricted use, modification, and distribution.
What is the context window size of Mixtral 8x22B?
Mixtral 8x22B supports a 64K token context window, enabling precise information recall from large documents.
Which languages does Mixtral 8x22B support?
It has native fluency in English, French, Italian, German, and Spanish.
How many parameters does Mixtral 8x22B have?
It has a total of 141B parameters, with only 39B active per token due to its sparse mixture-of-experts architecture.
Can Mixtral 8x22B be fine-tuned?
Yes, the base model is available for fine-tuning, making it an excellent basis for custom domain-specific applications.
Does Mixtral 8x22B support function calling?
Yes, it natively supports function calling, along with a constrained output mode, enabling application development and tech stack modernization at scale.