Activefastllm Open Source

Mistral Small 3

by Mistral AI· Released January 2025· Cutoff July 2024

Mistral Small 3 is a 24B-parameter dense language model optimized for low-latency, high-throughput inference, offering strong performance at a fraction of the cost of larger models. It is designed to be a fast, efficient alternative for tasks that do not require the full power of frontier models, and it is released under the Apache 2.0 license.

Official Site API Docs 🤗 Hugging Face

Input cost

Free (open source)

Output cost

Free (open source)

Context window

32K tokens

Max output

—

Modalities

text

Parameters

24B

License

Apache-2.0

Capabilities

Text GenerationCode GenerationFunction CallingStreamingJSON ModeMultilingual Support

Best For

Low-latency, cost-effective text and code generation tasks where speed and efficiency are prioritized over maximum capability.

Strengths

Very fast inference speed, suitable for real-time applications
Competitive performance for its size, outperforming many larger models on certain benchmarks
Apache 2.0 license allows free use, modification, and distribution
Efficient architecture with 24B parameters, balancing quality and resource usage

Limitations

Smaller context window (32K tokens) compared to some competitors
Not a frontier model; may lag behind larger models on complex reasoning or specialized tasks
Limited to text-only input; no native vision or multimodal capabilities
Knowledge cutoff is relatively early (July 2024), so recent events may be missing

Use Cases

Real-time chatbots and virtual assistants

Code generation and autocompletion in IDEs

Text summarization and content generation

Data extraction and structured output generation

Customer support automation

Educational tools and tutoring systems

Lightweight on-device or edge deployment

Improvements Over Previous Model

New model in the Mistral Small family, replacing Mistral Small 2 (22B) with a 24B parameter count
Apache 2.0 license, making it fully open-source compared to previous proprietary models
Improved latency and throughput for real-time applications
Enhanced performance on coding and reasoning benchmarks relative to its size

Back to all models

Strengths

Very fast inference speed, suitable for real-time applications

Competitive performance for its size, outperforming many larger models on certain benchmarks

Apache 2.0 license allows free use, modification, and distribution

Efficient architecture with 24B parameters, balancing quality and resource usage

Limitations

Smaller context window (32K tokens) compared to some competitors

Not a frontier model; may lag behind larger models on complex reasoning or specialized tasks

Limited to text-only input; no native vision or multimodal capabilities

Knowledge cutoff is relatively early (July 2024), so recent events may be missing

Improvements Over Previous Model

New model in the Mistral Small family, replacing Mistral Small 2 (22B) with a 24B parameter count

Apache 2.0 license, making it fully open-source compared to previous proprietary models

Improved latency and throughput for real-time applications

Enhanced performance on coding and reasoning benchmarks relative to its size