Umranz/Qwen3.8-27B-heretic-v2

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Umranz/Qwen3.8-27B-heretic-v2 is a 27 billion parameter causal language model, based on the Qwen3.8 architecture, with a native context length of 262,144 tokens. This version is a decensored variant of Umranz/Qwen3.8-27B-heretic, further processed using Heretic v1.4.0. It is a vision-language model designed for advanced agentic tasks, excelling in coding, professional work, research, and multimodal understanding of images and videos.

Loading preview...

Model Overview

Umranz/Qwen3.8-27B-heretic-v2 is a 27 billion parameter causal language model, building upon the Qwen3.8 architecture. This specific version is a decensored iteration of Umranz/Qwen3.8-27B-heretic, created using Heretic v1.4.0. It features a native context length of 262,144 tokens, extensible up to 1,000,000 tokens using YaRN scaling, and includes native support for image and video understanding.

Key Capabilities

  • Enhanced Agentic Performance: Demonstrates strong autonomous planning and improved handling of environmental feedback, leading to more reliable end-to-end task completion across coding, office work, and professional tasks.
  • Multimodal Understanding: Offers native vision-language capabilities, processing images and videos, including STEM diagrams, documents, and hour-scale video content.
  • Flexible Thinking Control: Incorporates a default 'thinking mode' with adjustable reasoning depth (reasoning_effort) and preserved reasoning context (preserve_thinking) for complex problem-solving.
  • Superior Coding & Agentic Benchmarks: Outperforms several comparable models in agentic terminal coding (73.0 on Terminal Bench 2.1), agentic coding (61.7 on SWE-bench Pro, 42.2 on DeepSWE 1.1), and long-horizon office work (70.7 on CoWorkBench).
  • Multimodal Agentic Intelligence: Achieves high scores in computer use (84.3 on OSWorld-Verified), browser use (64.8 on WebArena-Verified), and mobile use (81.9 on AndroidWorld).

When to Use This Model

  • Complex Agentic Workflows: Ideal for applications requiring advanced autonomous planning, multi-step task completion, and robust handling of environmental feedback.
  • Coding and Software Engineering: Particularly strong for agentic coding tasks, repo-level code generation, and software engineering benchmarks.
  • Multimodal Applications: Suitable for tasks involving image and video analysis, visual reasoning, document intelligence, and multimodal tool use.
  • Long Context Processing: Benefits from a large native context window and support for YaRN scaling for ultra-long text inputs.

Finetunes
1 models