Taichu model releases

Taichu (ZiDong Taichu / 紫东太初) is a Chinese multimodal foundation-model series originating from the Institute of Automation, Chinese Academy of Sciences (CASIA) and the Wuhan Artificial Intelligence Research Institute, first launched in 2021. It focuses on multimodal understanding across text, image, audio and video. First tracked model is ZDTaichu5.0-9B (Sep 2026), published under the TaichuAI Hugging Face handle. This page collects the lab's model releases, lifecycle events, source links, and model metadata in one crawlable record.

Models

Labs

Open

Recent

All models·Latest releases·Open-weight models·Changelog·Home

Most recent in this set

Sep 24, 2026

ZDTaichu5.0-9B

Taichu

Sort

1 model

ZDTaichu5.0-9B

Available

TaichuOpen weights

ZDTaichu5.0-9B is an open-weights spatial vision-language model from the Taichu (ZiDong Taichu) series, published ~2026-09-24 under the NVIDIA Open Model License (retaining Qwen3.5 Apache-2.0 and third-party notices). ~10B parameters combining a Qwen3.5-9B decoder with an NVIDIA C-RADIOv4-H vision encoder. Multimodal understanding: accepts text, single or multiple images and video at any resolution and outputs text, with a focus on spatial reasoning. Up to 128K-token context. Reported benchmarks include ViewSpatial 62.5, MMSI 47.2, MindCube-tiny 78.3, TAU2 87.7, Claw-Eval 71.4 and IFEval 93.7. A derivative built on a Qwen backbone, added for parity with other tracked multimodal-understanding VLMs.

Dense10B131K ctxSep 24, 2026