trohrbaugh/Qwen3.6-35B-A3B-heretic-v2

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 25, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

This is a 35.1 billion parameter Qwen3.6-35B-A3B causal language model, developed by Qwen and decensored by trohrbaugh using Heretic v1.2.0. It features a 32768 token context length and is optimized for agentic coding, handling frontend workflows and repository-level reasoning with enhanced stability and real-world utility. The model also supports thinking preservation, allowing it to retain reasoning context from historical messages for streamlined iterative development.

Loading preview...

Model Overview

trohrbaugh/Qwen3.6-35B-A3B-heretic-v2 is a 35.1 billion parameter causal language model based on the Qwen3.6-35B-A3B architecture, developed by Qwen and subsequently decensored by trohrbaugh using Heretic v1.2.0. This model is designed with a native context length of 32,768 tokens, extensible up to 1,010,000 tokens using YaRN scaling techniques.

Key Capabilities

  • Decensored Output: This version has been modified to reduce refusals, achieving 0/100 refusals compared to 99/100 in the original model.
  • Agentic Coding: Excels in handling frontend workflows and repository-level reasoning, offering improved fluency and precision for coding tasks.
  • Thinking Preservation: Features an option to retain reasoning context from historical messages, which streamlines iterative development and reduces overhead.
  • Multimodal Support: Capable of processing text, image, and video inputs, making it suitable for diverse applications.
  • High Performance: Demonstrates strong performance across various benchmarks, particularly in coding agent tasks (e.g., 73.4 on SWE-bench Verified, 51.5 on Terminal-Bench 2.0) and vision language tasks (e.g., 92.8 on MMBench EN-DEV-v1.1).

Use Cases

This model is particularly well-suited for:

  • Agent-based Development: Ideal for building AI agents that require advanced coding capabilities and contextual reasoning.
  • Complex Coding Tasks: Its enhanced agentic coding features make it effective for intricate programming challenges, including frontend development and repository-wide analysis.
  • Multimodal Applications: Suitable for applications requiring understanding and generation based on combined text, image, and video inputs.
  • Iterative Development Workflows: The thinking preservation feature benefits scenarios where maintaining reasoning context across multiple interactions is crucial.

Finetunes
1 models