Sulat.com
AI models
$10 off the fastest DeepSeek V4.1 Flash, Kimi K3 and GLM 5.3 from Synthetic
LLM Gateway logo

Model details

Mistral Small 4 (Mistral AI)

Mistral Small 4 represents a major release in the Mistral Small family, designed as a single hybrid system that combines instruction following, extended reasoning, and coding ability. Built on a Mixture-of-Experts-style architecture with 119 billion total parameters and 6.5 billion active per inference, the model is engineered to handle demanding analytical workloads without requiring users to switch between separate models for different tasks. Released under the Apache 2.0 license as version v26.03 on March 16, 2026, it is open-weight, making it accessible for self-hosting, fine-tuning, and enterprise customization while still being available through managed API endpoints.

In practical terms, Mistral Small 4 is best suited for complex problem solving, mathematical reasoning, planning, and multi-step analysis where chain-of-thought generation adds clear value. The model accepts both text and image inputs, enabling multimodal workflows such as document or diagram understanding paired with structured text reasoning. Its large context window supports long-form analysis, code repositories, and agentic pipelines with function calling and structured output. Trade-offs to consider are higher inference latency and elevated compute cost relative to smaller Mistral variants, so it fits best where reasoning depth matters more than raw response speed, such as research assistance, code generation on hard problems, and orchestrated agent workflows.

LLM Gatewaymistral/mistral-small-2603mistral-small

Quick Info

Powered by
Provider
LLM Gateway
Model key
mistral/mistral-small-2603
Release date
Mar 16, 2026
Last updated
Mar 16, 2026
Knowledge cutoff
2025-06
Input modalities
Output modalities
Capabilities

Cost

Input token cost
$0.15
Output token cost
$0.60

Limits

Output tokens
256,000 tokens
Context window
262,144 tokens

Transparent token rates

Compare Mistral Small 4 (Mistral AI) pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Mistral Small 4 (Mistral AI)

LLM Gateway

Coverage

Mozilla and Mistral AI announced on 16 September 2026 that Mistral Small 4, a 119-billion-parameter open-weight model, is being added to Firefox Smart Window beta in the US and Canada. The deal also expands Smart Window beta access and French-language support to Firefox users in France. Mozilla's post describes Mistral Small 4 as a new AI model for Smart Window users, while Mistral's announcement frames it as one option among several in the AI models menu. Both companies emphasized open source, user choice, and European sovereignty in the partnership launch communications.

LLM Gateway

Coverage

Mistral Small 4 is listed among Mistral's current open-weight options alongside Mistral Large 3 and Mistral Medium 3.5 in a 2026 open-weight model comparison guide. The guide positions Mistral Small 4 for practical local and enterprise deployments, distinct from frontier closed models aimed at the hardest tasks. The piece frames open-weight models as part of a broader portfolio strategy rather than wholesale replacements for Claude, GPT, or Gemini. While Mistral Small 4 is grouped with Mistral's open-weight releases for accessible local use, the guide does not enumerate specific benchmark scores, context window sizes, or licensing terms for the 2603 variant. It notes that some contenders still rely on vendor-reported figures rather than independent verification, so teams should weigh evidence quality alongside workload fit. No release date or first-party Mistral documentation is linked from the excerpt for Mistral Small 4 specifically.

LLM Gateway

CoverageRelease Notes

Mistral AI released Mistral Small 4 in March 2026 as its first model to unify instruction following, reasoning, multimodal understanding, and agentic coding into a single deployment target. It consolidates roles previously split across Mistral Small, Magistral, Pixtral, and Devstral. The model is a Mixture-of-Experts architecture with 128 experts and 4 active per token, totaling 119B parameters with 6B active per token. It supports a 256k context window, handles text and image inputs with text output, and targets general chat, coding, agentic tasks, and complex reasoning workloads.

Videos about Mistral Small 4 (Mistral AI)

More models around Mistral Small 4 (Mistral AI)