Mistral Small 4 logo

Mistral Small 4 Review

Hybrid AI model combining reasoning, coding, and vision in one open-source system

Large Language Models (LLMs)New
No ratings yet
Visit Mistral Small 4
View Alternatives
Mistral Small 4 screenshot

Mistral Small 4 is a Large Language Models (LLMs) tool. Hybrid AI model combining reasoning, coding, and vision in one open-source system. Best for software developers and engineers, data scientists and analysts and scientists and researchers.

6 key features6+ alternatives →

About Mistral Small 4

Mistral Small 4 is a 119B-parameter open-source language model that unifies instruction following, configurable reasoning, multimodal vision, and agentic coding capabilities. It features a 256K context window and Apache 2.0 licensing.

Key Features

**Unified Model Architecture.** Mistral Small 4 combines instruction following, reasoning, vision understanding, and coding capabilities in a single model, removing the need to switch between specialized models for different tasks.
**Configurable Reasoning Depth.** The reasoning_effort parameter lets you toggle between fast responses for everyday tasks and deep step-by-step reasoning for complex problems, all within the same model endpoint.
**Mixture of Experts Design.** Uses 128 expert networks with only 4 active per token, giving you 119B total parameters with just 6.5B active parameters per inference, keeping costs low while maintaining quality.
**Multimodal Input Support.** Accepts both text and image inputs for document parsing, visual analysis, and data extraction tasks, with a 256K token context window for handling large documents and codebases.
**Open-Source Apache 2.0 License.** Fully open-source with commercial use rights, fine-tuning capabilities, and redistribution permissions, allowing you to self-host on your own infrastructure without restrictions.
**Cost-Efficient Pricing.** Available at $0.15 per million input tokens and $0.60 per million output tokens through the API, with 40% lower latency and 3x higher throughput compared to the previous version.

Frequently Asked Questions

Mistral Small 4 unifies reasoning, coding, and vision capabilities in a single model instead of requiring separate specialized models. It also features configurable reasoning depth per request, letting you choose between fast responses and deep reasoning without switching models.

Yes, Mistral Small 4 is released under Apache 2.0 license with full commercial use rights. You can self-host it, but you'll need significant GPU infrastructure—minimum 4x NVIDIA H100 or 2x H200 GPUs. The full weights are available on Hugging Face.

Through the Mistral API, Mistral Small 4 costs $0.15 per million input tokens and $0.60 per million output tokens. If you self-host under the Apache 2.0 license, there are no licensing fees, but you'll need to cover your own infrastructure costs.

Mistral Small 4 supports a 256K token context window, which is roughly 190,000 words of usable context. This is large enough to handle full codebase analysis, lengthy legal documents, and multi-session conversations without aggressive chunking.

User Reviews

Similar Tools

View all →