OpenIntelligenceNet/Qwen-3.8-27B-Heretic

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 14, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Qwen3.8-27B is a 27 billion parameter causal language model developed by Qwen, built upon the Qwen3.5 architecture. This native vision-language model supports image and video understanding, excelling in coding, professional work, research, and long-horizon agentic tasks. It features flexible thinking control and a 32,768 token context length, extensible up to 1,000,000 tokens with YaRN scaling.

Loading preview...

Qwen3.8-27B: Advanced Multimodal Agentic Capabilities

Qwen3.8-27B is the latest iteration in the Qwen open-model family, a 27 billion parameter causal language model with native vision-language understanding. It builds on the Qwen3.5 architecture, delivering significant improvements across various domains.

Key Capabilities & Enhancements

  • Multimodal Understanding: Native support for interpreting images and videos, from STEM diagrams to hour-scale video content.
  • Enhanced Agent Execution: Features stronger autonomous planning and improved handling of environment feedback, leading to more reliable completion of complex, multi-step tasks.
  • Flexible Thinking Control: Includes configurable 'thinking mode' with adjustable reasoning depth (reasoning_effort - xhigh, medium, low) and retention of reasoning context (preserve_thinking).
  • Broad Compatibility: Offers wider support for popular harnesses and development tools, simplifying integration.
  • Extended Context Length: Natively supports 32,768 tokens, extensible up to 1,000,000 tokens using YaRN scaling techniques for ultra-long texts.

Benchmark Highlights

Qwen3.8-27B demonstrates strong performance, often leading its class in agentic and multimodal benchmarks:

  • Coding: Achieves 61.7 on SWE-bench Pro (agentic coding) and 79.0 on QwenSWEBench (software engineering).
  • Agentic Tasks: Scores 70.7 on CoWorkBench (long-horizon office work) and 33.4 on JobBench (professional job tasks).
  • Multimodal Agentic Intelligence: Leads with 84.3 on OSWorld-Verified (computer use), 64.8 on WebArena-Verified (browser use), and 81.9 on AndroidWorld (mobile use).
  • General Multimodal Intelligence: Achieves 94.6 (with CI) on MathVision (visual math problem solving) and 85.6 (with CI) on BabyVision (general visual reasoning).

Recommended Use Cases

This model is particularly well-suited for applications requiring:

  • Complex Agentic Workflows: Ideal for tasks demanding autonomous planning, tool use, and reliable multi-step execution across various digital environments (desktop, web, mobile).
  • Multimodal Content Analysis: Excellent for processing and understanding information from both text and visual inputs, including images, diagrams, and long-form videos.
  • Advanced Coding & Software Engineering: Strong performance in agentic coding, repo-level code generation, and software engineering tasks.
  • Scientific & Professional Reasoning: Capable in scientific reasoning, multidisciplinary problem-solving, and handling professional job-related tasks.

Finetunes
1 models