Change feed / 2026-08-07
HY3 stays under-review on data handling — but the evidence we cited was a pricing page
Our data-handling entry for HY3 cited the Tencent Cloud TokenHub page as its evidence. That page is marketing and pricing; it contains no data-handling terms at all, and never did. Tencent Cloud's actual privacy policy has now been located and read in full: across roughly 92,000 characters it does not mention Hunyuan, model training, generative AI, or large language models once. The grade is unchanged — under-review remains correct — but the reason is now documented and checkable rather than inferred. There is no first-party statement in either direction on whether TokenHub prompts train the model. Worth stating plainly: a top-ten model by usage whose provider publishes nothing about training use is itself the finding, and self-hosting the Apache 2.0 weights is the only way to remove the question.
Sources (1)
- Tencent Cloud Privacy Policy (general; silent on model training)
“We will retain your personal information as described below (except if otherwise required to be retained by applicable law).”
Plain
ModelRiskIndex. "HY3 stays under-review on data handling — but the evidence we cited was a pricing page." Change feed, 2026-08-07. https://modelriskindex.com/changes/mri-hy3-data-handling-evidence-corrected
BibTeX
@misc{mri-2026-08-07,
title = {HY3 stays under-review on data handling — but the evidence we cited was a pricing page},
author = {{ModelRiskIndex}},
year = {2026},
note = {Change feed, 2026-08-07},
url = {https://modelriskindex.com/changes/mri-hy3-data-handling-evidence-corrected}
}Please cite the dated entry rather than the site root. Every assessment here is a point-in-time judgment bound to evidence retrieved on a specific date — an undated citation asserts something the data does not.
