xiry.

AI · AI model

Pixtral Large

Mistral AI's 124B parameter open-weights multimodal vision and language model

From the Xiry library · Updated 3 October 2026

Visit mistral.ai ↗

Pixtral Large is a 124-billion-parameter multimodal model built by Mistral AI, combining a 123B decoder with a 1B vision encoder to understand text and high-resolution images. It is designed to handle visual math, document comprehension, and chart analysis at frontier-grade levels. Users can interact with it directly through Mistral's Le Chat web assistant or integrate it via API and self-hosted model weights.

What it's for

  • Analyze complex charts, infographics, and technical diagrams
  • Extract data and answer questions from dense multi-page documents
  • Solve visual math problems and code visual UI mockups
  • Build custom multimodal AI agents with web search and reasoning

Features

  • 124B multimodal architecture (123B decoder and 1B vision encoder)
  • High-resolution image and multi-image processing capabilities
  • Strong visual reasoning benchmarked on DocVQA, MathVista, and MM-MT-Bench
  • Direct web access via Mistral's Le Chat interface
  • Available as open weights and via Mistral's API platform

Pricing

Freemium. Free to use via Le Chat; pay-as-you-go via Mistral API

Details

Type
AI model
Category
AI
Topic
Multimodal Models
Open source
Yes
Platforms
Web, API, Self-hosted
For
Developers, researchers, and technical teams needing open-weight frontier multimodal AI
Alternative to
GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro
Tags
multimodal, vision, llm, mistral, open weights, ai

Why it was saved

  • @5aitec, 23 November 2024: Competes with and outperforms frontier multimodal models like GPT-4o on visual math, chart analysis, and complex document comprehension benchmarks.
  • @5aitec, 23 November 2024: Offers a ChatGPT alternative equipped with real-time web search with citations, image generation, 300+ page document summarization, and a custom agent builder.

Compare

Related

Updated