DevJac/Qwen3.8-27B-heretic
Qwen3.8-27B-heretic is a 27 billion parameter causal language model from the Qwen3.8 series, developed by Qwen. This model is a native vision-language model capable of understanding images and videos, and is specifically enhanced for agentic tasks, coding, and professional work. It features flexible thinking control and an extended context length of 262,144 tokens, extensible up to 1,000,000 tokens, making it suitable for complex, multi-step problem-solving.
Loading preview...
Model Overview
Qwen3.8-27B is a 27 billion parameter causal language model from the Qwen3.8 series, building upon the Qwen3.5 architecture. It is designed as a native vision-language model, capable of processing both image and video inputs, alongside text. The model demonstrates significant advancements in agentic capabilities, coding performance, and handling professional and research tasks.
Key Capabilities
- Multimodal Understanding: Natively supports image and video inputs, enabling analysis of STEM diagrams, documents, and hour-long videos.
- Enhanced Agent Execution: Features improved autonomous planning and more robust handling of environmental feedback for reliable, end-to-end task completion.
- Flexible Thinking Control: Offers adjustable reasoning depth (
reasoning_effortwithxhigh,medium,lowsettings) and preserves reasoning context across turns (preserve_thinking). - Extended Context Window: Natively supports a context length of 262,144 tokens, extensible up to 1,000,000 tokens using YaRN scaling techniques.
- Strong Coding Performance: Achieves leading scores on agentic coding benchmarks like SWE-bench Pro (61.7) and DeepSWE 1.1 (42.2), and competitive coding on LiveCodeBench v6 (90.3).
- Agentic Multimodal Intelligence: Excels in tasks requiring computer use (OSWorld-Verified 84.3), browser use (WebArena-Verified 64.8), and multimodal tool use (ClawEval-MM Pass@3 57.4).
Good For
- Complex Agentic Workflows: Ideal for applications requiring autonomous planning and execution across various domains, including long-horizon office tasks and professional job tasks.
- Advanced Code Generation and Debugging: Particularly strong in agentic coding scenarios, repo-level code generation, and software engineering tasks.
- Multimodal Data Analysis: Suitable for tasks involving visual reasoning, scientific chart analysis, document intelligence, and real-world perception from images and videos.
- High-Context Applications: Benefits use cases requiring processing and generating very long texts or detailed multimodal inputs, leveraging its extended context capabilities.