Activefastllm Open Source

Mixtral 8x7B

by Mistral AI· Released December 2023· Cutoff December 2023

Mixtral 8x7B is a sparse mixture-of-experts (MoE) model that achieves high performance with low latency by activating only 12.9B parameters per token. It is Mistral's first open-weight MoE model, offering strong multilingual capabilities and a 32K token context window.

Official Site API Docs 🤗 Hugging Face 📄 Research Paper

Input cost

Free (open source)

Output cost

Free (open source)

Context window

32K tokens

Max output

—

Modalities

text

Parameters

46.7B total (12.9B active per token)

License

Apache-2.0

Capabilities

MultilingualCode GenerationFunction CallingStreamingJSON Mode

Best For

High-quality text generation and code tasks with efficient inference via MoE architecture.

Strengths

Excellent performance-to-compute ratio due to MoE design
Strong multilingual support (English, French, Italian, German, Spanish)
Competitive with larger models like GPT-3.5 on many benchmarks
Open-weight availability for self-hosting and fine-tuning

Limitations

Not suitable for tasks requiring very long context beyond 32K tokens
May not match the reasoning depth of larger dense models like GPT-4
Limited to text-only input; no vision or multimodal capabilities
Requires careful memory management due to 46.7B total parameters

Use Cases

Chatbots and virtual assistants

Code generation and assistance

Content creation and translation

Summarization and document analysis

Educational tutoring and Q&A

Data extraction and parsing

Fine-tuning for domain-specific applications

Improvements Over Previous Model

First MoE model from Mistral, enabling efficient inference with only 12.9B active parameters
Context window increased from 8K (Mistral 7B) to 32K tokens
Outperforms Llama 2 70B on most benchmarks while being 6x faster in inference
Supports multiple languages natively, unlike many models focused on English
Open-weight release under Apache 2.0 license, allowing broad usage

Back to all models

Activefastllm Open Source

Mixtral 8x7B

by Mistral AI· Released December 2023· Cutoff December 2023

Official Site API Docs 🤗 Hugging Face 📄 Research Paper

Input cost

Free (open source)

Output cost

Free (open source)

Context window

32K tokens

Max output

—

Modalities

text

Parameters

46.7B total (12.9B active per token)

License

Apache-2.0

Capabilities

MultilingualCode GenerationFunction CallingStreamingJSON Mode

Best For

High-quality text generation and code tasks with efficient inference via MoE architecture.

Strengths

Excellent performance-to-compute ratio due to MoE design
Strong multilingual support (English, French, Italian, German, Spanish)
Competitive with larger models like GPT-3.5 on many benchmarks
Open-weight availability for self-hosting and fine-tuning

Limitations

Not suitable for tasks requiring very long context beyond 32K tokens
May not match the reasoning depth of larger dense models like GPT-4
Limited to text-only input; no vision or multimodal capabilities
Requires careful memory management due to 46.7B total parameters

Use Cases

Chatbots and virtual assistants

Code generation and assistance

Content creation and translation

Summarization and document analysis

Educational tutoring and Q&A

Data extraction and parsing

Fine-tuning for domain-specific applications

Improvements Over Previous Model

First MoE model from Mistral, enabling efficient inference with only 12.9B active parameters
Context window increased from 8K (Mistral 7B) to 32K tokens
Outperforms Llama 2 70B on most benchmarks while being 6x faster in inference
Supports multiple languages natively, unlike many models focused on English
Open-weight release under Apache 2.0 license, allowing broad usage

Back to all models