As Simon Willison notes, the sheer size of Hy3—nearly 600GB for the full model—makes its 'open-source' label somewhat academic for most developers. The more significant development may be its temporary free availability on OpenRouter, which serves as a clever distribution and stress-testing tactic. This release appears to be part of a broader pattern where Chinese tech giants, facing export restrictions, are leveraging open-source communities for global benchmarking and influence. The key watchpoint is whether the claimed efficiency gains translate into tangible adoption outside Tencent's own ecosystem, or if it remains a trophy model for the leaderboard.
Tencent releases Hy3, a large Chinese open-source MoE model, reports say
The 295B-parameter model is said to outperform similar-sized rivals and is available for free testing on OpenRouter.
AIpressr commentary on an article originally published by Simon Willison.
For informational purposes only. AI-assisted commentary may contain errors. full disclaimer ↓hide ↑
This is AIpressr's editorial commentary on a report originally published by another outlet — it is opinion, not the original reporting, and not an endorsement by or affiliation with that outlet. Follow the linked source for the underlying facts. Editorial & AI disclosure.
Editor's Take
Simon Willison reports on Tencent's release of Hy3, a massive open-source mixture-of-experts model. The immediate question is who, beyond researchers with significant GPU resources, can actually run a model of this scale, and what its availability signals about China's open-source AI strategy. In our view, the push to rival flagship models with fewer active parameters is a notable technical claim, but the real-world utility across '50+ products' remains to be independently verified.
“Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team.”
Our analysis
Have AI news to share?
Submit your release →Publisher or subject of this story? Object to this commentary or request a correction →
