GEAR: 10× faster autoregressive image generation Tencent Hunyuan's new method jointly trains VQ tokenizers and AR generators end-to-end, beating LlamaGen-REPA with a novel dual read-out. All tokenizers are on Hugging Face.

Minimax H3 is the first *open* video model I've seen that outperforms HunyuanVideo 1.5 - which is actually an impressive accomplishment for @TencentHunyuan to have held that throne for so long. For a while now I've been sad that the image/video frontier seemed to be getting much more closed than the text LLM frontier. H3 is a very strong step back in the open direction. This was generated in ~30 min on my AMD laptop:

Introducing Hyra-1.0, the first version of Hunyuan Research Agent. 💡💡💡 Built to recursively improve solutions for performance-driven research and engineering tasks. Explore our demos in AI4AI, AI4Science, and AI4Fun:

📢 Hy4 Preview Is Now Live on API! Developed by Tencent Hunyuan, Hy4 Preview is a 770B sparse MoE open-weights model activating 49B parameters per token, demonstrating strong performance (2.99/4.00) in internal expert evaluations on 203 engineering tasks. Official API access is now available! 👉 Try now: 🔗 Learn more:

🙏 Thank you all for the incredible love and support! Our latest Tencent Hunyuan translation models are on fire on Hugging Face: 🥰Hy-MT2-1.8B ranks #1# 🥰Hy-MT2-30B-A3B ranks #4# on the open-source model trending leaderboard, with over 7K downloads already! To make it even easier for everyone, we’ve launched the Tencent Hy Translation WeChat mini-program, built on Hy-MT2. It supports voice input and offline translation, plus powerful customization of translation styles and instructions — delivering results that better match your expectations and feel far more practical. Try it out and share your feedback with us — we’d love to hear from you! Models on HF: GitHub: #HyMT2# #TencentHunyuan# #OpenSource#

A new open weight Chinese model has hit the timeline!! This time it’s from Tencent’s Hunyuan team with Hy4 preview. It’s a massive MoE with 770B total parameters / 49B active per token, a 1M context window, and the weights are released under Apache 2.0. Important caveat - Tencent says this is still an early Hy4 checkpoint with more pretraining + post training to come. I actually like how honest their benchmark sheet is. They show plenty of places where the model is still behind despite its size. Hy4 gets 85.4 on Terminal Bench 2.1, basically right in the frontier cluster, and jumps from Hy3’s 28.0 -> 64.3 on DeepSWE. But it’s still behind Kimi K3 at 74.0 and Claude Opus 5 at 74.7 there. On ProgramBench it gets 17.5 vs Claude’s 39.5, SWE Atlas Refactoring 53.3 vs 60.0, and Humanity’s Last Exam 43.4 vs 53.2. It’s a 49B active open weight model that is competitive for its size on a bunch of hard coding/agent benchmark. One thing Tencent also reported on GitHub is they ran a 163 person internal blind eval across 203 engineering tasks where Hy4 slightly beat GLM 5.3 and Kimi K3, which is pretty interesting.

We've just merged its first community contribution! 🚀 6 new models benchmarked and submitted via open PR: DeepSeek, Kimi K2, Hunyuan, Nova Pro, Command A, Jamba and two new countries on the map. 24 models total, from 6 countries. And because trust is the whole point: community runs are displayed flagged as "pending re-verification" until we reproduce them ourselves. Open methodology means anyone can contribute, and nobody, including us, gets taken on faith. Thank you bricepirard-spec. Who's next?

TENCENT $700: AI catalysts last couple days. "Workbuddy" AI agent combined with new HY3 model launch seeing usage growth & news of ex OpenAI researcher becoming new head of "Hunyuan" AI division. Also, obviously a cheap stock and certain parallels w/ MAG7 rotation.

Every demo in this video came out of a single prompt. No follow ups, nothing fixed by hand. I ran 168 of them through an agent harness, one shot each, and picked 20 for the cut. Median 7 minutes per demo. The neon tunnel burned 98 056 output tokens to produce 5 KB of code, it thinks a lot more than it writes. 68 of the 168 came out working in a browser. Most of the misses were my own harness capping a reply at 64 000 tokens, not the model. This is Hy4 preview, Tencent Hunyuan just open sourced it. 770B total, 49B active, 1M+ context, third flagship they ship in 6 months. The part i find more interesting than the size is how they built it. They co-designed it next to their own products instead of throwing the weights over the wall, with people who actually do software, games, finance and security. In their internal blind test, 163 experts across 203 engineering tasks, it scored 2.99/4 against Kimi K3 at 2.94 and GLM 5.3 at 2.92. Their numbers, not mine. Price is the part that will annoy some people. $0.834/M in, $2.501/M out, $0.042/M cache hits. Hy4 preview is free on WorkBuddy for 2 weeks right now if you want to poke at it yourself: @TencentHunyuan @TencentAI_News @WorkBuddy_AI