Ep 649 GitHub July 13, 2026 0:53 w/ Pippa & Tyler

tencent/Hy3 · Hugging Face

Tencent releases Hy3, a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters, open-sourced under Apache 2.0 on Hugging Face.

Read the source → Plain-text transcript →

Embed this episode

Paste this on any site — the player is a self-contained iframe with no cookies or trackers.

<iframe src="https://sandrise.io/exploring-next/embed/649"
  width="100%" height="180" style="max-width:640px;border:0;border-radius:12px;overflow:hidden"
  title="Exploring Next — Episode 649 audio player"
  loading="lazy" allow="autoplay" referrerpolicy="strict-origin-when-cross-origin"></iframe>

Embed & API docs →

Script Llama 4 Scout Voice Inworld TTS 1.5 Mini

Transcript

Pippa Hey Tyler, have you seen the new Hy3 model from Tencent? It's a 295B-parameter MoE model that's been open-sourced on Hugging Face.

Tyler Yeah, I checked it out. The interesting thing is that it's a Mixture-of-Experts model with 21B active parameters. That's quite a unique approach.

Pippa Right. And it's been open-sourced under Apache 2.0, which is great for the community. I wonder what kind of applications people will build on top of it.

Tyler Definitely. The benchmarks look promising too. It achieves 90.4 on the GPQA benchmark, which is impressive.

Pippa That's really interesting. I read that Tencent rebuilt their infrastructure from scratch to train this model. That's a significant investment.

Tyler Yeah, that's a big deal. It shows they're serious about their AI research efforts. I'm curious to see how this model performs in real-world applications.