Qwen/Qwen3.8-27B

Qwen3.8-27B: Advanced Multimodal Agentic Capabilities

Qwen3.8-27B is the latest and most capable generation in the Qwen open-model family, a 27 billion parameter causal language model with a native vision encoder. Building on the Qwen3.5 architecture, this model delivers significant improvements across various domains, particularly in agentic tasks and multimodal understanding.

Key Capabilities & Enhancements

• Comprehensive Core Capabilities: Substantial gains in coding, professional work, research, and long-horizon agentic tasks.
• Robust Agent Execution: Features stronger autonomous planning and improved handling of environment feedback, leading to more reliable end-to-end task completion.
• Flexible Thinking Control: Operates in a default 'thinking mode' with adjustable reasoning depth (reasoning_effort - xhigh, medium, low) and preserved reasoning context (preserve_thinking). This can be disabled for direct responses.
• Native Vision-Language Understanding: Supports understanding of both images and videos, including STEM diagrams, documents, and hour-scale video content.
• Extended Context Length: Natively supports a context length of 262,144 tokens, extensible up to 1,000,000 tokens using techniques like YaRN for ultra-long texts.

Performance Highlights

Qwen3.8-27B demonstrates strong performance across various benchmarks, often outperforming comparable models:

• 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: Excels in visual math problem solving (94.6 with CI on MathVision) and general visual reasoning (85.6 with CI on BabyVision).

Ideal Use Cases

This model is particularly well-suited for developers and researchers focused on:

• Complex Agentic Workflows: Building agents that require robust planning, execution, and multi-step task completion.
• Multimodal Applications: Developing applications that integrate image and video understanding with advanced language processing.
• Advanced Coding & Software Engineering: Tasks involving agentic coding, repo-level code generation, and software engineering problem-solving.
• Long-Context Processing: Scenarios requiring analysis of extensive documents, codebases, or long video content.