SwarmDo/SwarmDo-A2-merged

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 23, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

SwarmDo/SwarmDo-A2-merged is a 27 billion parameter model based on Qwen/Qwen3.6-27B, developed by SwarmDo. This model uniquely combines coding capabilities with render-verified visual-coding, allowing it to fix bugs and generate code from images. It is designed for direct serving with vLLM, offering both text-based code generation and image-to-code functionality within a single model.

Loading preview...

SwarmDo-A2-merged Overview

SwarmDo-A2-merged is a 27 billion parameter model built upon the Qwen/Qwen3.6-27B architecture, developed by SwarmDo. This version comes with the combined coding and render-verified visual-coding adapter already merged, including the vision tower, making it ready for direct serving with vLLM without an additional merge step.

Key Capabilities

  • Unified Code & Vision: Integrates both text-based code generation and visual-coding (image-to-code) capabilities into a single model.
  • Execution-Verified Bug Fixing: Proven to fix real bugs with execution verification.
  • Render-Verified Visual Coding: Capable of writing code from images, with its output verified through rendering.
  • No Performance Degradation: Measured proof that combining these two skills does not degrade individual performance (the "churn gate").
  • vLLM Ready: Optimized for direct serving with vLLM, supporting OpenAI-compatible chat endpoints for both text and image inputs.

When to Use This Model

This model is ideal for applications requiring a powerful, unified solution for:

  • Automated Code Generation: Generating code based on textual prompts or problem descriptions.
  • Visual-to-Code Translation: Creating code directly from visual inputs, such as UI mockups or diagrams.
  • Bug Fixing & Code Refinement: Leveraging its execution-verified bug-fixing capabilities.
  • Integrated Development Environments: As a backend for tools that need both code understanding and visual interpretation.