AGW06/Qwen3.6-35B-A3B

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:2Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 14, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Qwen3.6-35B-A3B is a 35.1 billion parameter causal language model developed by Qwen, featuring a Mixture of Experts (MoE) architecture with 3 billion activated parameters and a native context length of 262,144 tokens, extensible up to 1,010,000 tokens. This model is optimized for agentic coding, offering enhanced frontend workflow handling and repository-level reasoning. It also supports thinking preservation from historical messages, streamlining iterative development and improving decision consistency in agent scenarios, and is vision-capable, supporting image and video inputs.

Loading preview...

Qwen3.6-35B-A3B Overview

Qwen3.6-35B-A3B is a 35.1 billion parameter causal language model from the Qwen series, featuring a Mixture of Experts (MoE) architecture where 3 billion parameters are activated. It boasts a native context length of 262,144 tokens, which can be extended up to 1,010,000 tokens using YaRN scaling techniques. This model is designed to provide a more intuitive and productive coding experience, building on community feedback from the Qwen3.5 series.

Key Capabilities

  • Agentic Coding: Significantly improved handling of frontend workflows and repository-level reasoning, making it highly effective for complex coding tasks.
  • Thinking Preservation: A unique feature that retains reasoning context from historical messages, enhancing decision consistency and reducing redundant reasoning in iterative development.
  • Multimodal Support: Capable of processing both image and video inputs, making it suitable for a wide range of vision-language tasks.
  • High Performance: Achieves strong results across various benchmarks, including coding agent tasks (e.g., 73.4 on SWE-bench Verified, 51.5 on Terminal-Bench 2.0), general agent tasks, knowledge, STEM, and vision language benchmarks (e.g., 81.7 on MMMU, 92.8 on MMBench).
  • Flexible Deployment: Compatible with popular inference frameworks like Hugging Face Transformers, vLLM, SGLang, and KTransformers, with specific optimizations for each.

Good For

  • Developers requiring advanced coding assistance: Especially for agentic workflows, frontend development, and repository-level code understanding.
  • Applications needing long context windows: With native support for 262K tokens and extensibility to over 1M tokens, it's ideal for processing extensive codebases or lengthy documents.
  • Multimodal AI applications: Its vision capabilities make it suitable for tasks involving image and video analysis alongside text.
  • Iterative development and agent systems: The thinking preservation feature is particularly beneficial for maintaining context and consistency in multi-turn interactions and complex agentic tasks.

Finetunes
1 models