OpenVisTool Models

This checkpoint is part of OpenVisTool, a family of open multimodal models trained for tool-augmented visual reasoning. The models can interleave reasoning, visual tool calls, tool observations, and final answers to solve visually complex tasks.

The models are supervised fine-tuned on OpenVisTool-42K, which contains 42,048 tool-use trajectories. The release includes four checkpoints:

Checkpoint	Base model
qwen3vl_8b_openvistool_42k	Qwen3-VL-8B-Instruct
qwen35_4b_openvistool_42k	Qwen3.5-4B
qwen35_9b_openvistool_42k	Qwen3.5-9B
qwen35_27b_openvistool_42k	Qwen3.5-27B