Qwen3.5-MoE vision architecture
The captured configuration identifies Qwen3_5MoeForConditionalGeneration with model type qwen3_5_moe and Transformers support.
Open Source Model Profile · Qwen
Qwen3.6-35B-A3B
Qwen3.6-35B-A3B is a 35.95B-parameter Qwen MoE image-text-to-text model from Qwen. According to the model card, it activates 3B parameters with 262,144-token native context.
- Publisher
- Qwen
- Task
- image-text-to-text
- Model type
- qwen3_5_moe
- License
- apache-2.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Qwen3.6-35B-A3B is published by Qwen as an image-text-to-text model. The captured configuration identifies Qwen3_5MoeForConditionalGeneration with model type qwen3_5_moe, and Safetensors metadata reports about 35.95B parameters. According to the model card, it is the first open-weight Qwen3.6 variant focused on agentic coding, with 35B total and 3B activated parameters.
Recorded capabilities
35B total with 3B activated
Safetensors metadata reports 35,951,822,704 parameters; according to the model card, the MoE has 256 experts with 8 routed plus 1 shared expert activated.
262K native context
According to the model card, context length is 262,144 natively and extensible up to 1,010,000 tokens, with YaRN supported across Transformers, vLLM, KTransformers, and SGLang.
Agentic coding positioning
According to the model card, the publisher positions Qwen3.6 for agentic coding, frontend workflows, repository-level reasoning, and tool calling.
Documented serving and sampling setup
According to the model card, the publisher documents SGLang and vLLM server commands, multi-token prediction options, text-only serving, and thinking versus instruct sampling parameters.
Use cases in the source record
- Agentic coding workflows including frontend tasks and repository-level reasoning with the publisher-documented tool-calling support.
- Long-context image-text deployment with the publisher-documented 262,144-token serving commands across vLLM, SGLang, and KTransformers.
Limitations and unknowns
- No Ethen-measured benchmark results were extracted.
- According to the captured page text, third-party leaderboard figures for GPQA, AIME 2026, SWE-bench Verified, and ExtractBench appear alongside the card; they are snapshot claims, not Ethen-verified results.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
Source and provenance
Source: Qwen/Qwen3.6-35B-A3B
Captured: Unknown. Processed: 2026-09-07T19:34:36.496777+00:00.
Qwen3.6-35B-A3B This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format. These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, KTransformers, etc. Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience. Qwen3.6 Highlights This release delivers substantial upgrades, particularly in Agent…
F001F002F003F004F005F006F007F008F010F012F013F014F015F018F020F022F026F027F033