Tooony133/Qwen-3.6-27B-BiggerFish

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 19, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Tooony133/Qwen-3.6-27B-BiggerFish is a 27 billion parameter causal language model with a vision encoder, developed by Qwen. This model prioritizes stability and real-world utility, excelling in agentic coding tasks, including frontend workflows and repository-level reasoning. It features a native context length of 262,144 tokens, extensible up to 1,010,000 tokens, and introduces thinking preservation for streamlined iterative development. Qwen-3.6-27B is designed for developers seeking a responsive and productive coding experience with advanced multimodal capabilities.

Loading preview...

Qwen-3.6-27B: Enhanced Agentic Coding and Multimodal Capabilities

Qwen-3.6-27B is a 27 billion parameter causal language model with a vision encoder, developed by Qwen, focusing on stability and real-world utility. This release introduces significant upgrades, particularly in its agentic coding capabilities and thinking preservation features.

Key Capabilities and Differentiators

  • Agentic Coding Excellence: The model demonstrates enhanced fluency and precision in handling frontend workflows and repository-level reasoning. Benchmarks show strong performance in coding agent tasks like SWE-bench Verified (77.2), SWE-bench Pro (53.5), and Terminal-Bench 2.0 (59.3).
  • Thinking Preservation: A novel feature allows the model to retain reasoning context from historical messages, which streamlines iterative development, reduces overhead, and improves decision consistency in agent scenarios.
  • Multimodal Understanding: As a vision-capable model, it supports image and video inputs, performing well across various vision language benchmarks including MMMU (82.9), MathVista mini (87.4), and VideoMME (87.7).
  • Extended Context Length: Natively supports a context length of 262,144 tokens, with extensibility up to 1,010,000 tokens using RoPE scaling techniques like YaRN, making it suitable for ultra-long text processing.
  • Robust Performance: Achieves competitive scores in knowledge (MMLU-Redux 93.5, C-Eval 91.4) and STEM & Reasoning benchmarks (GPQA Diamond 87.8, AIME26 94.1).

Use Cases and Recommendations

Qwen-3.6-27B is ideal for developers building applications that require:

  • Advanced Code Generation and Refactoring: Especially for complex frontend and repository-level tasks.
  • Intelligent Agents: Leveraging its tool-calling capabilities and thinking preservation for more consistent and efficient agentic workflows.
  • Multimodal AI Applications: Integrating image and video understanding with complex reasoning.
  • Long-Context Processing: For tasks involving extensive documentation, codebases, or conversational histories.

For optimal performance, specific sampling parameters are recommended for general tasks, precise coding, and instruct modes. The model operates in a thinking mode by default, generating reasoning steps before final responses, which can be disabled if direct output is preferred.