Tooony133/Qwen-3.6-27B-CrystalRun

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 6, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Qwen3.6-27B is a 27 billion parameter causal language model developed by Qwen, featuring a vision encoder and a native context length of 262,144 tokens, extensible up to 1,010,000 tokens. This model is specifically optimized for agentic coding, enhancing frontend workflows and repository-level reasoning, and introduces thinking preservation for iterative development. It excels in complex coding tasks, mathematical problem-solving, and multimodal understanding across images and videos.

Loading preview...

Overview

Qwen3.6-27B is a 27 billion parameter causal language model from Qwen, building upon the Qwen3.5 series with a focus on stability and real-world utility. It incorporates a vision encoder and supports a substantial native context length of 262,144 tokens, which can be extended up to 1,010,000 tokens using RoPE scaling techniques like YaRN.

Key Capabilities

  • Agentic Coding: Significantly improved handling of frontend workflows and repository-level reasoning, making it highly effective for complex coding tasks.
  • Thinking Preservation: Features an innovative option to retain reasoning context from historical messages, which streamlines iterative development and can reduce token overhead.
  • Multimodal Understanding: Capable of processing and understanding both image and video inputs, demonstrated by strong performance across various vision language benchmarks.
  • Extended Context: Natively supports a very long context window, crucial for complex problems and maintaining conversational flow over extended interactions.

Benchmarks & Performance

The model shows competitive performance across a range of benchmarks, particularly in:

  • Coding Agent: Achieves 77.2 on SWE-bench Verified and 59.3 on Terminal-Bench 2.0, outperforming Qwen3.5-27B and Gemma4-31B in several coding agent tasks.
  • Knowledge & Reasoning: Scores 86.2 on MMLU-Pro and 93.5 on MMLU-Redux, alongside strong results in STEM and reasoning tasks like GPQA Diamond (87.8) and AIME26 (94.1).
  • Vision Language: Demonstrates robust multimodal capabilities with scores like 82.9 on MMMU and 97.0 on VlmsAreBlind, indicating strong performance in STEM & Puzzle, General VQA, Document Understanding, Spatial Intelligence, and Video Understanding tasks.

Good for

  • Developers requiring advanced coding assistance: Especially for agentic workflows, frontend development, and repository-level code analysis.
  • Applications needing long-context reasoning: Ideal for tasks where maintaining extensive conversational or code context is critical.
  • Multimodal AI solutions: Suitable for scenarios involving image and video analysis alongside text, such as visual question answering or video content understanding.
  • High-throughput inference: Optimized for deployment with frameworks like SGLang, vLLM, and KTransformers for efficient serving.