wesjos/mario-html-v2-4B

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 14, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

wesjos/mario-html-v2-4B is a 4.5 billion parameter language model developed by wesjos, based on Qwen/Qwen3.5-4B, with a native context length of 262,144 tokens. This model is uniquely designed for text-to-HTML generation, and can also handle text-plus-image to HTML, producing structured HTML output with predefined CSS classes. Its primary use case is generating dynamic, styled web UI components directly from natural language prompts, differing from typical LLMs that output plain text or markdown.

Loading preview...

Overview

wesjos/mario-html-v2-4B is a 4.5 billion parameter model built upon the Qwen/Qwen3.5-4B base, featuring a substantial native context length of 262,144 tokens. This model specializes in generating HTML output directly from text or text-plus-image inputs, making it distinct from models that primarily produce plain text or Markdown. Version 2 of this model reduces the presence of HTML tags within its Chain-of-Thought (CoT) reasoning traces, aiming for cleaner internal planning.

Key Capabilities

  • Direct HTML Generation: Produces structured HTML with predefined CSS classes, suitable for web UI components.
  • Multimodal Input: Supports both text-to-HTML and text-plus-image-to-HTML transformations.
  • Structured Output Contract: Adheres to a specific set of HTML components and class names (e.g., ai-view, summary, card, grid, metric-group, callout, table, code-box).
  • Integrated WebUI: Comes with a bundled WebUI for demonstration, featuring dual-stream rendering (collapsible <think> traces and live HTML), image paste, and multiple theme presets.
  • Reasoning Traces: Can display internal <think> traces (plain-text planning) before generating the final HTML, offering insight into its generation process.

Good For

  • Developers needing to generate styled web UI elements directly from natural language.
  • Prototyping web interfaces or dynamic content quickly without manual HTML/CSS coding.
  • Applications requiring structured, component-based HTML output rather than free-form text.
  • Exploring models that integrate reasoning with structured output generation.

Finetunes
2 models