🔥 @TencentHunyuan Hy4 preview just cited a Zhihu post with a surprisingly simple challenge to DeepSeek’s mHC: What if the doubly stochastic matrix is overcomplicating things—and Identity actually works better? Today, let’s revisit the cited post, “Your DeepSeek mHC May Not Need the ‘m’,” by Zhihu contributor 涮月亮的谪仙人. After training Qwen3 1.7B and 8B dense models from scratch on 150B tokens, the experiments found: 💡 Identity HC > mHC > mHC-lite > orthogonal mHC In other words: simply setting H_res = Identity beat the Sinkhorn-Knopp constrained version. Here’s why 👇 1️⃣ What does mHC actually change? Standard Transformers have one residual stream. Hyper-Connections (HC) expand it to multiple parallel streams: 🔹 H_pre reads from the streams 🔹 H_post writes back to them 🔹 H_res mixes information between them DeepSeek’s mHC constrains H_res to a doubly stochastic matrix via Sinkhorn-Knopp, helping preserve norms and stabilize propagation. But the experiments suggest a simpler question: Do we need H_res mixing at all? 2️⃣ mHC seems to learn something close to Identity anyway For a single layer, the learned H_res is already close to Identity: diagonal ≈ 0.96, off-diagonal ≈ 0.01 But multiply H_res across many layers, and it gradually collapses toward a uniform 0.25 matrix. So each layer may look almost like Identity, while their cumulative effect becomes uniform mixing. The simplest fix? Just set H_res = I. 3️⃣ Identity preserves stream semantics With Identity, each residual stream stays where it is: Stream 0 stays Stream 0. Stream 1 stays Stream 1. No repeated reshuffling, no cumulative mixing, and Iᴸ = I. H_pre and H_post also no longer need to track where each stream has been repeatedly moved—they simply learn where to read and where to write. 4️⃣ But cross-stream communication still happens Setting H_res = I does not isolate the streams. The projection that generates H_pre and H_post already sees all residual streams, making H_pre input-dependent. So information can still be dynamically aggregated across streams before Attention/MLP and written back afterward. H_res isn’t the only mechanism for cross-stream interaction. 5️⃣ Why might Sinkhorn hurt? Repeated products of positive doubly stochastic matrices tend toward uniform mixing. In the Qwen3-1.7B experiment, after 56 HC modules, the minimum singular value of the accumulated H_res product reached just: 9.2 × 10⁻¹⁸ By ~10 layers, the four streams were already approaching the same 0.25 uniform mixture. Sinkhorn also comes with extra cost: 20 iterations, backward recomputation, extra parameters, and approximation error. Identity has none of these—and preserves the residual signal exactly. 6️⃣ More sophisticated alternatives didn’t win either The experiments also tested mHC-lite, softmax-weighted convex combinations, and orthogonal variants using Cayley/Givens transforms. The observed ranking remained: Identity HC > mHC > mHC-lite > orthogonal mHC The simplest design won. 💡 The takeaway DeepSeek’s mHC uses sophisticated manifold constraints to stabilize Hyper-Connections. But these experiments suggest that H_res itself may not need to be learned or mixed at all. Sometimes the best manifold constraint is the most boring one: H_res = I. Or, as the original post puts it: Maybe DeepSeek’s mHC doesn’t need the “m.” 😆 👉 Read the full Zhihu post for the training curves, mathematical analysis, H_res visualizations, and implementation details: #DeepSeek# #mHC# #Hunyuan# #Tencent# #LLM# #Transformer# #AIResearch# #AI#

Hy4 Preview from @TencentHunyuan is live on AI Gateway. • 𝚝𝚎𝚗𝚌𝚎𝚗𝚝/𝚑𝚢𝟺-𝚙𝚛𝚎𝚟𝚒𝚎𝚠 • Open source MoE at 770B params, 49B active, 1M context window

Tencent Hy3 from @TencentHunyuan is free on @OpenRouter through July 21. 295B MoE, 256K context, built for coding, reasoning, agents and reliable tool use. Try it in OpenClaw today: openclaw models set openrouter/tencent/hy3:free

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 Hy ASR 3.0 preview, a speech recognition model from @TencentHunyuan built to understand, not just transcribe What we improved: - Cleaner on real-world audio: dialects, code-switching, and long-form content with less error accumulation over time - Context-aware correction: homophones and ambiguous phrases get resolved from context, not guessed in isolation - Hotword injection: drop in brand names, people, domain terms without retraining. Lowers integration cost for niche use cases - Built for noisy rooms: whisper, background noise, tricky acoustic conditions stay stable

Tried Hy4 preview through WorkBuddy and gave it a pretty vague prompt for a first person roller coaster ride, front row seat, make it feel fast, that was basically it. No detailed spec, no reference video, just a rough idea in a sentence. What came back actually held up. Steep drop that felt like the bottom dropping out, banked turns that leaned the right way, scenery rushing by close enough that you register the speed instead of just watching it happen. For something built off a loose prompt like that, I did not expect it to nail the feel this well. Coding and 3D rendering have clearly gotten a real upgrade here, and this took one pass to look this good. Free to try in @WorkBuddy_AI for the next two weeks if you want to mess with it yourself. @TencentHunyuan @TencentAI_News

Hy4 Preview def wins this 3D sim of the ISS in orbit contest! the best frontend output in comparison to Qwen 3.8 Flash, GLM 5.3, and DeepSeek V4 Flash Vision what a crazy week, and GLM 5.3 open weight is coming tmr 🤯 @TencentHunyuan

Most AI platforms make you choose between model quality, flexibility, and cost. is taking a different approach. Its current 100% FREE model lineup gives users access to three very different models, each suited to a different kind of workload. DeepSeek-V4-Flash from @deepseek_ai is built for speed and high-concurrency use. It’s a strong option when you need fast reasoning across coding, agentic workflows, and other demanding tasks. DeepSeek-V4-Flash-Vision-Exp adds multimodal capabilities. Instead of working only with text, it brings vision into the workflow, making it useful for image understanding and vision-based applications. Then there’s Tencent Hy3 from @TencentHunyuan, a general-purpose model with strong Chinese-language capabilities and support for function calling. That makes it particularly interesting for applications that need more than simple text generation. The important part is that these aren’t three versions of the same thing. They give builders different capabilities to experiment with: V4 Flash → speed & reasoning V4 Flash Vision Exp → multimodal intelligence Tencent Hy3 → general-purpose AI & function calling And is making this lineup available 100% FREE. That changes the equation for developers, AI enthusiasts, and Web3 builders. You can test an idea without immediately worrying about inference costs. You can compare models against real workloads. You can experiment with agents, coding, vision, and tool-based applications before deciding what deserves a larger production setup. This is especially interesting at a time when AI development is moving toward agentic systems, where models need to reason, understand different types of information, call tools, and complete multi-step tasks. Lowering the cost of experimentation means more people can actually build. And has already crossed 1 trillion tokens in total throughput in just 5 days, showing how quickly users are putting the infrastructure to work. If you’ve been looking for a place to experiment with capable LLMs without starting with a bill, this is worth checking out. Three models. Different strengths. 100% free access. Explore #TRONEcoStar# @justinsuntron #TRON# @BAI_AGI

12 Chinese AI models made their World Cup predictions. ⚽ The World Cup Round of 32 lineup is out.🔥 The first leaderboard is in.👇 Tencent Hy went 29/32, ranking #1# among the 12 models after the Round of 32 was set. Now comes the harder test: whose predictions survive the knockout stage? #TencentHunyuan# #AI# #WorldCup#