As usual, disclosure first: I'm on the team that built this. Our MiroThinker-1.7-deepresearch and 1.7-mini-deepresearch API went live, a deep research agent built on Qwen3 MoE (30B total, 3B active for mini). Weights on HuggingFace: huggingface.co/miromind-ai/MiroThinker-1.7 Posting here because the