Qwen 3.6 MoE SOTA Agentic Coding 73.4% SWE-Bench Local Boom
Key Questions
What is Qwen3.6-35B-A3B?
Qwen3.6-35B-A3B is an open-source MoE model with 35B total and 3B active parameters, achieving 73.4% on verified SWE-Bench. It supports multimodal inputs, 256k context, Apache 2.0 license, and runs locally on MacBook/4090 at 180 tok/s.
How does Hermes Agent work with Qwen 3.6?
Hermes Agent with Qwen 3.6 enables self-improving AI, available on Ollama, DashScope, OpenRouter. It scores 37% on MCPMark for agentic tasks.
What production setups use Qwen 3.6?
OpenClaw setups optimize Qwen 3.6 for production workflows, focusing on latency and cost efficiency.
Qwen3.6-35B-A3B 73.4% SWE Verified 3B active/35B multimodal 256k ctx Apache2.0 local MacBook/4090 180tok/s; Hermes Agent self-improve Ollama/DashScope/OpenRouter MCPMark37%; OpenClaw setups for prod workflows latency/cost.
Sources (2)
Updated Apr 18, 2026