Sponsored by Byond Boundrys Consulting - Empowering Ideas, Delivering Results
Mistral AI Advanced

Mistral Small 4

Mistral Small 4 is a current 256K-context multimodal hybrid model with Apache 2.0 weights and $0.15/M input, $0.60/M output API pricing.

Multimodal ModelsTextImage Freemium
In plain English

What is this model and why does it matter?

Mistral Small 4 is a 119B-total, 6.5B-active multimodal hybrid model combining instruct, reasoning and coding in a 256K context window.

General assistantsReasoningCoding agentsDocument understandingMultimodal analysisSelf-hosting
Model overview

Mistral Small 4: features, use cases and important details

Mistral Small 4 is Mistral AI’s March 2026 hybrid model combining instruction following, reasoning, coding and multimodal understanding.

Verified model facts

Mistral documents 119B total parameters, about 6.5B active parameters, 256K context, text and image input, Apache 2.0 licensing, function calling and structured outputs.

Current status

It is an active GA model with current API access and open weights.

Best fit

It fits agents, coding, document understanding, multimodal assistants and cost-sensitive enterprise chat.

Limitations

The full model is still large to self-host and, like other generative systems, can produce incorrect reasoning or visual interpretations.

Get started

How to use this model

  1. Call Mistral with model mistral-small-2603 or deploy the open weights.
  2. Provide text and optional image input.
  3. Set reasoning effort based on latency/quality needs.
  4. Use function calling or structured outputs when required.
  5. Benchmark hosted vs self-hosted cost.
Copy and try

Example prompts

  • Analyze this document and image together.
  • Reason through this coding problem with high effort.
  • Return the answer as structured JSON and call tools if needed.
Capabilities

What it can do

  • 256K context
  • Multimodal input
  • Configurable reasoning
  • Coding
  • Function calling
  • Structured outputs
  • Open weights
Best for

Practical use cases

  • Agents
  • Coding
  • Document Q&A
  • Multimodal assistants
  • Enterprise chat
Pricing

What does it cost?

$0.15 per 1M input tokens and $0.60 per 1M output tokens on Mistral's current API.

Input$0.15 / 1M tokens
Output$0.60 / 1M tokens
Simple summaryCurrent Mistral API pricing is $0.15/M input and $0.60/M output; open Apache 2.0 weights also support self-hosting.

What stands out

  • Apache 2.0
  • Low API price
  • 256K context
  • Reasoning plus coding plus vision
  • Open weights

Things to consider

  • Large 119B total parameter footprint
  • Self-hosting still requires substantial hardware
Limitations

Important restrictions and trade-offs

  • Can hallucinate
  • Large-context and image workflows increase compute cost
  • Self-host deployment complexity remains significant
SimplifyAITools verdict

Our editorial take

One of the strongest current-value Mistral options because it combines open weights, multimodality, reasoning and low API pricing.

References

Primary sources

  1. Open source 1 ↗
  2. Open source 2 ↗
  3. Open source 3 ↗
Keep exploring

Related AI models