coder3101/Qwen3.8-27B-heretic

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The coder3101/Qwen3.8-27B-heretic is a 27 billion parameter, decensored version of the Qwen3.8-27B causal language model, built using Heretic v1.2.0. This model features native vision-language understanding, flexible thinking control, and enhanced agentic capabilities for complex, multi-step tasks. It excels in coding, professional work, research, and long-horizon agentic tasks, supporting a native context length of 262,144 tokens, extensible up to 1,000,000 tokens.

Loading preview...

Overview

This model, coder3101/Qwen3.8-27B-heretic, is a 27 billion parameter, decensored variant of the Qwen3.8-27B model, created using Heretic v1.2.0. It is part of the Qwen3.8 series, which represents the latest generation in the Qwen open-model family, offering significant advancements over previous versions. The model is a native vision-language model, capable of understanding both images and videos, and is designed for robust performance in complex, multi-step tasks.

Key Capabilities

  • Decensored: Modified to reduce refusals, with 33/100 refusals compared to the original model's 92/100.
  • Enhanced Agent Execution: Features stronger autonomous planning and improved handling of environment feedback, leading to more reliable task completion.
  • Vision-Language Understanding: Natively supports image and video understanding, including STEM diagrams, documents, and hour-scale videos.
  • Flexible Thinking Control: Includes a 'thinking mode' that can be enabled/disabled, with adjustable reasoning depth (reasoning_effort) and preserved reasoning context (preserve_thinking).
  • High Performance: Demonstrates strong benchmark results in agentic coding (e.g., 61.7% on SWE-bench Pro, 79.0% on QwenSWEBench), agentic multimodal intelligence (e.g., 84.3% on OSWorld-Verified), and general text/multimodal tasks.
  • Extended Context Length: Supports a native context length of 262,144 tokens, extensible up to 1,000,000 tokens using YaRN scaling techniques.

Good For

  • Agentic Applications: Ideal for tasks requiring autonomous planning, complex problem-solving, and reliable multi-step task completion, especially in coding and professional domains.
  • Multimodal Understanding: Suitable for applications that need to process and reason over both text and visual inputs (images and videos).
  • Coding and Software Engineering: Excels in various coding benchmarks, including agentic terminal coding, repo-level code generation, and software engineering tasks.
  • Research and Professional Work: Offers comprehensive improvements across these domains, benefiting from enhanced reasoning and long-horizon task capabilities.