Qwen3.6-35B-A3B drops as open weights

Efficiency, not raw scale: another Qwen release that tries to make “local model” and “serious agent” compatible in the same package.

• –A 35B total / 3B active MoE setup is the sweet spot for throughput, latency, and hardware cost.
• –If the Apache 2.0/open-weights framing holds, teams can test, fine-tune, and deploy without vendor lock-in.
• –This keeps pressure on other open-weight model families by pushing capability into a more runnable size class.
• –The most likely use case is coding assistants, tool-using agents, and other workflows where speed matters as much as quality.