Tencent Hy(@TXhunyuan)

We want to address recent reports regarding agent activity. The activity originated from evaluatio...

Tencent Hy (@TencentHunyuan) — Fri, 16 Oct 2026 00:06:22 GMT

We want to address recent reports regarding agent activity. 

The activity originated from evaluations of third-party models, not Tencent Hy. During these evaluations, some agents exhibited unintended web-browsing behavior, but our review found no evidence of deliberate targeting or coordination. We take these findings seriously. We are strengthening network access controls and agent monitoring.

We are committed to building safe and responsible AI. Thanks to the community for their feedback.

💬31🔄42❤️532👀45987📊124

AMAZING😍

Tencent Hy (@TencentHunyuan) — Wed, 14 Oct 2026 06:57:16 GMT

AMAZING😍

GMI Cloud@gmi_cloud

Announcing the HY Image Challenge winners! 

We received so many creative submissions, and picking just 3 winners was hard.

Together with the @TencentHunyuan team, we selected 3 winners and rounded up some of our favorite creations.

Check out their work below 👇

Your browser does not support the video tag.

💬1🔄2❤️47👀5517📊6

more 🔛 https://t.co/XsFS9gaRpQ

Tencent Hy (@TencentHunyuan) — Fri, 09 Oct 2026 02:36:57 GMT

more 🔛 github.com/Tencent-Hunyua…Q

Tencent Hy@TencentHunyuan

New Research: We are releasing ExplorationBench, a benchmark for measuring how AI systems explore.

Scientific discovery begins where known problems end: a system has to frame hypotheses, design experiments, and learn from the results. Evaluating this is hard. Genuinely new answers cannot be checked quickly, and in familiar domains a model can simply recall what it has seen.

Addressing this challenge, researchers from Tencent Hy, Fudan University, and Tsinghua University built verifiable Alien Worlds. Their rules are executable, so every answer is checked exactly, and they conflict with familiar knowledge, so recall alone cannot solve the tasks.

🔹 Two sandboxes: AlienCode (31 hidden rule changes, 70 tasks) and AlienLogic (24 patched inference rules, 70 theorems)
🔹 A flawed manual, four rounds of self-designed probes, and closed-book tests after every round
🔹 Every answer graded by an interpreter or a proof checker, with no LLM judge

What we found across 10 frontier AI systems:
1️⃣ Getting feedback is more effective than thinking alone. No AlienCode run starts above 15.7%; after four rounds the best reaches 89.0%, while the same turns without feedback stay at 0.5–11.0%.
2️⃣ Designing the experiments matters. Replaying a system's own best probes gives it exactly the same evidence, yet in AlienCode 9 of 10 systems do worse than when they chose the probes themselves.
3️⃣ Knowing a rule is not using it. Even when every required rule is stated correctly, tasks are solved only 73.4% of the time.
4️⃣ One score hides a lot. The same system under the same budget ended anywhere from 5.7% to 79.0%, and rankings barely transfer between the two worlds.

CL-bench asked whether models can learn from context. ExplorationBench asks whether they can discover the rules themselves.

📄 Paparxiv.org/abs/2609.30199JyxQ
🌐 Website & leaderboexplorationbench.comCfc2s
📝 explorationbench.com/blog/H7ZnQH
💻 Code (coming github.com/Tencent-Hunyua…lgErkcl

💬0🔄1❤️45👀6672📊6

New Research: We are releasing ExplorationBench, a benchmark for measuring how AI systems explore. ...

Tencent Hy (@TencentHunyuan) — Wed, 30 Sep 2026 13:29:01 GMT

New Research: We are releasing ExplorationBench, a benchmark for measuring how AI systems explore.

Scientific discovery begins where known problems end: a system has to frame hypotheses, design experiments, and learn from the results. Evaluating this is hard. Genuinely new answers cannot be checked quickly, and in familiar domains a model can simply recall what it has seen.

Addressing this challenge, researchers from Tencent Hy, Fudan University, and Tsinghua University built verifiable Alien Worlds. Their rules are executable, so every answer is checked exactly, and they conflict with familiar knowledge, so recall alone cannot solve the tasks.

🔹 Two sandboxes: AlienCode (31 hidden rule changes, 70 tasks) and AlienLogic (24 patched inference rules, 70 theorems)
🔹 A flawed manual, four rounds of self-designed probes, and closed-book tests after every round
🔹 Every answer graded by an interpreter or a proof checker, with no LLM judge

What we found across 10 frontier AI systems:
1️⃣ Getting feedback is more effective than thinking alone. No AlienCode run starts above 15.7%; after four rounds the best reaches 89.0%, while the same turns without feedback stay at 0.5–11.0%.
2️⃣ Designing the experiments matters. Replaying a system's own best probes gives it exactly the same evidence, yet in AlienCode 9 of 10 systems do worse than when they chose the probes themselves.
3️⃣ Knowing a rule is not using it. Even when every required rule is stated correctly, tasks are solved only 73.4% of the time.
4️⃣ One score hides a lot. The same system under the same budget ended anywhere from 5.7% to 79.0%, and rankings barely transfer between the two worlds.

CL-bench asked whether models can learn from context. ExplorationBench asks whether they can discover the rules themselves.

📄 Paparxiv.org/abs/2609.30199JyxQ
🌐 Website & leaderboexplorationbench.comCfc2s
📝 explorationbench.com/blog/H7ZnQH
💻 Code (coming github.com/Tencent-Hunyua…lgErkcl

💬14🔄45❤️430👀70086📊116

3 tracks. 3 winners. Go and Create. 🥇

Tencent Hy (@TencentHunyuan) — Wed, 30 Sep 2026 08:36:31 GMT

3 tracks. 3 winners. Go and Create. 🥇

GMI Cloud@gmi_cloud

Hy Image 3.5 preview is FREE on GMI Cloud for 7 days 

Generate + edit images up to 2K. Bring up to 5 reference images. Build posters, product shots, game art, and more.  

And we’re turning the free week into a challenge:  

3 tracks. 3 winners. $1,800 in cash + credits.  

Start creating free 👇

Your browser does not support the video tag.

💬1🔄2❤️54👀8009📊8

live on GMI Cloud 🔥

Tencent Hy (@TencentHunyuan) — Thu, 24 Sep 2026 08:58:33 GMT

live on GMI Cloud 🔥

GMI Cloud@gmi_cloud

Hy Image 3.5 preview is now live on GMI Cloud @TencentHunyuan

$0.024 per image 
8.8x cheaper than GPT Image 2 
5.6x cheaper than Nano Banana Pro   

Create now 👇

💬1🔄1❤️51👀7944📊7

🚀 Tencent Hy Translation just landed.Powered by Hy-MT2. 33 languages. 5 Chinese minority languages...

Tencent Hy (@TencentHunyuan) — Thu, 24 Sep 2026 07:15:39 GMT

🚀 Tencent Hy Translation just landed.Powered by Hy-MT2.

33 languages. 5 Chinese minority languages & dialects.
Voice. Photo. Full offline — on-device, no network required.
Already live in 12 countries and regions.
Travel, drive, work or read abroad. Accurate. Natural. Always available.
⏬ ⏬⏬ apps.apple.com/us/app/tencent…s

💬34🔄59❤️644👀51309📊174

⚡️ As LLM reinforcement learning scales to larger GPU clusters and more training data, training effi...

Tencent Hy (@TencentHunyuan) — Thu, 24 Sep 2026 03:15:39 GMT

⚡️ As LLM reinforcement learning scales to larger GPU clusters and more training data, training efficiency becomes a first-order concern.

Our new research revisits classical critical-batch-size theory and extends it to online LLM RL, where the model generates its own training data and rollout generation and training scale differently.

Across GRPO and PPO, we find that learning-rate retuning can preserve learning per response over a bounded range of batch sizes.

On fixed hardware, scaling up the batch size improves PPO generation-stage throughput by up to 2.29×, while our best measured GRPO configuration reaches the same validation target in 29% less time. 🚀

Read the full research:hy.tencent.ai/research/100116q

💬10🔄8❤️194👀14897📊48

⚡️ As LLM reinforcement learning scales to larger GPU clusters and more training data, training effi...

Tencent Hy (@TencentHunyuan) — Thu, 24 Sep 2026 02:29:57 GMT

⚡️ As LLM reinforcement learning scales to larger GPU clusters and more training data, training efficiency becomes a first-order concern.

Our new research revisits classical critical-batch-size theory and extends it to online LLM RL, where the model generates its own training data and rollout generation and training scale differently.

Across GRPO and PPO, we find that learning-rate retuning can preserve learning per response over a bounded range of batch sizes.

On fixed hardware, scaling up the batch size improves PPO generation-stage throughput by up to 2.29×, while our best measured GRPO configuration reaches the same validation target in 29% less time. 🚀

Read the full research:hy-ai.woa.com/research/10011…5

💬0🔄1❤️20👀836📊3

Free for two weeks！Hy Image3.5 preview is live on OnSolo. Short drama character sheets. Full-motion ...

Tencent Hy (@TencentHunyuan) — Wed, 23 Sep 2026 06:56:34 GMT

Free for two weeks！Hy Image3.5 preview is live on OnSolo. Short drama character sheets. Full-motion video game assets. Keyframes. Characters hold across every episode. Edits refine instead of restarting.

OnSoloAI@OnSoloAI

Hy Image3.5 preview is on OnSolo. Exclusive. 5 refs. 2K. 2 weeks Members free.

Your browser does not support the video tag.

💬12🔄6❤️136👀10862📊30

ComfyUI ✖️ Hy Image3.5 preview

Tencent Hy (@TencentHunyuan) — Wed, 23 Sep 2026 02:40:02 GMT

ComfyUI ✖️ Hy Image3.5 preview

ComfyUI@ComfyUI

Hy Image3.5 preview is now available in ComfyUI.

Professional-grade image generation, +30% win rate in human eval vs Hy Image3.0

→ Text to image and Image to image in one model, up to 2K
→ Multilingual text, symbols, and small print that render correctly
→ Cinematic, comic, commercial photography, and illustration styles
→ Identity and product features that hold through scene, outfit, and style changes

Your browser does not support the video tag.

💬2🔄3❤️85👀7788📊12

ComfyUI ✖️ Hy image3.5 preview

Tencent Hy (@TencentHunyuan) — Wed, 23 Sep 2026 02:27:02 GMT

ComfyUI ✖️ Hy image3.5 preview

ComfyUI@ComfyUI

Hy Image3.5 preview is now available in ComfyUI.

Professional-grade image generation, +30% win rate in human eval vs Hy Image3.0

→ Text to image and Image to image in one model, up to 2K
→ Multilingual text, symbols, and small print that render correctly
→ Cinematic, comic, commercial photography, and illustration styles
→ Identity and product features that hold through scene, outfit, and style changes

Your browser does not support the video tag.

💬0🔄1❤️1👀455📊1

🎨 Hy Image3.5 preview is live in Miora. Edits that keep what already works — same canvas, your bran...

Tencent Hy (@TencentHunyuan) — Tue, 22 Sep 2026 13:24:19 GMT

🎨 Hy Image3.5 preview is live in Miora. Edits that keep what already works — same canvas, your brand rules already remembered. Free for two weeks.

Miora Design@Miora_Design

Hy Image3.5 preview is live on Miora. FREE through October 7.
Campaign visuals. Product scenes. Storyboards. Game concepts.
Up to 2K. Text-to-image and image-to-image.
Try it: miora.design

Your browser does not support the video tag.

💬8🔄9❤️182👀18683📊34

Share some genius cases generated by Hy Image3.5 preview. 😎

Tencent Hy (@TencentHunyuan) — Tue, 22 Sep 2026 09:10:13 GMT

Share some genius cases generated by Hy Image3.5 preview.  😎

Tencent Hy@TencentHunyuan

Hy Image3.5 preview is live. 🚀

Professional-grade image generation, +30% win rate in human eval vs Hy Image3.0
Both Text to image & Image to image available.
Up to 2K. Better Consistency.

API:console.tencentcloud.com/tokenhub/model…Q
Priced for everyone. $0.024 per image on Tencent Cloud API.
We only charge for what we generate — your reference images are free.

Two weeks free (Only in OnSolo and Miora):
Mioramiora.designO
OnSoloonsolo.aik

Try it and tell us where it breaks.

Your browser does not support the video tag.

💬10🔄7❤️146👀12529📊25

Sure you can edits images with this model.✔️

Tencent Hy (@TencentHunyuan) — Tue, 22 Sep 2026 09:01:59 GMT

Sure you can edits images with this model.✔️

Jiří Coufal@JiriCoufal77

@TencentHunyuan just generation? can it do also edits?

💬3🔄1❤️54👀8002📊8

The AI says the site is done. Then the homepage errors, the buttons overlap, and it added a login fl...

Tencent Hy (@TencentHunyuan) — Tue, 22 Sep 2026 08:53:39 GMT

The AI says the site is done. Then the homepage errors, the buttons overlap, and it added a login flow you never asked for. 🙃
Introduces WebCraftBench: agents actually use the live app, coverage-guided exploration finds what never got reached, then we score aesthetics, usability, and whether the original request was met.
On 197 human-validated pairs, it matches human preference 85.3% of the time.

Paper:arxiv.org/abs/2609.15387Z

💬10🔄13❤️209👀35454📊57

Hy Image3.5 preview is live. 🚀 Professional-grade image generation, +30% win rate in human eval vs...

Tencent Hy (@TencentHunyuan) — Tue, 22 Sep 2026 02:40:42 GMT

Hy Image3.5 preview is live. 🚀

Professional-grade image generation, +30% win rate in human eval vs Hy Image3.0
Both Text to image & Image to image available.
Up to 2K. Better Consistency.

API:console.tencentcloud.com/tokenhub/model…Q
Priced for everyone. $0.024 per image on Tencent Cloud API.
We only charge for what we generate — your reference images are free.

Two weeks free (Only in OnSolo and Miora):
Mioramiora.designO
OnSoloonsolo.aik

Try it and tell us where it breaks.

Your browser does not support the video tag.

💬41🔄103❤️1129👀113227📊293

Less size, same intelligence.

Tencent Hy (@TencentHunyuan) — Tue, 22 Sep 2026 02:37:49 GMT

Less size, same intelligence.

Zhihu Frontier@ZhihuFrontier

How Tencent Packed a 770B-Parameter Model into 214 GiB
Shrinking Hy4 preview's weights from roughly 1.5TB to 214 GiB is one challenge. Preserving useful capabilities and practical inference speed is another. How did @TencentHunyuan tackle both?
Zhihu contributor yghstill, a member of Tencent Hunyuan's quantization team, explains the engineering behind it. The parameter count remains 770B; the compression changes how those weights are represented.

Four weights, five bits
Sherry is the quantization algorithm, STQ1_0 the storage format, and MIX-STQ1_0 the mixed-precision allocation scheme.
Each group of four weights takes values from {-d, 0, +d}, with exactly one zero. Four zero positions multiplied by eight sign combinations gives 32 possible patterns, requiring five bits.
That is 1.25 bits per weight for the codes alone. Including a shared FP16 scale for every 256 weights brings STQ1_0 to 1.3125 bits per weight. The complete mixed-precision model averages about 2.38 bits per weight.

Allocate precision to specific weights, not whole layers
Hy4 preview has 77 MoE layers, each with 256 routed experts. The team concentrates aggressive compression on expert weights while selectively protecting other components.
For the experts' gate/up projections, MIX-STQ1_0 uses IQ2_XXS on 48 sensitive layers and STQ1_0 on 29 less sensitive layers.
The author reports that mixing lower and higher precision produces less error at the same average bit budget than uniformly choosing the intermediate IQ1_M format.

Layer sensitivity needs more than diagonal statistics
The author describes using the full Hessian, H = XXᵀ, to measure quantization sensitivity. Its off-diagonal terms capture correlations that diagonal-only imatrix scoring misses.
In the team's comparison, the two sensitivity rankings had a Spearman correlation of -0.115. The chosen layers did not follow a simple “deeper means more important” rule: precision was allocated greedily by error reduction per additional byte.

Fit the scale and choose the zero together
This is post-training quantization, without retraining. The encoder alternates between two decisions: fitting d with weighted least squares and choosing the zero position using imatrix-weighted error.
Zeroing the smallest-magnitude weight is not always the best choice. What matters is the additional weighted error introduced by making it zero.
Across 1,200 rows of real expert weights, three alternating rounds reduced weighted reconstruction error by roughly 90% compared with the original ternary encoder. This measures local weight reconstruction, not end-to-end model accuracy.

Compression must survive the runtime
The team implemented STQ1_0 CUDA kernels in a patched llama.cpp build. In its operator comparison, STQ1_0 ran roughly as fast as IQ1_M despite the lower bit width.
The author reports nearly unchanged MRCR retrieval performance and a small decline in math. Against UD-IQ1_M at a similar bit budget, the mixed-precision model led across the reported evaluations, including a gain of more than five points on MRCR.
The result comes from combining compact encoding, calibrated quantization, selective precision and usable inference kernels.

💬2🔄11❤️184👀10730📊25

🚀 EvolveScaler is here. Read a 40-day RPG log. Now answer one question: if you skip the mini-boss ...

Tencent Hy (@TencentHunyuan) — Tue, 15 Sep 2026 06:34:00 GMT

🚀 EvolveScaler is here.

Read a 40-day RPG log. Now answer one question: if you skip the mini-boss on Day 7, do you still beat the final boss?

The answer isn't in the log. You have to replay the world.

That's Information Evolution — records get retracted, corrected, backfilled. The world keeps changing after you read it.

So we build it backwards: define the world as an executable state machine, then render it into natural language. Code guarantees the logic. Language delivers the mess.

➡️ 117 prototypes. 159 question operators. 5 difficulty tiers. Up to ~1,200 events per sample.

➡️ 14 frontier models, hardest tier: median avg@5 falls to 11.3.

➡️ Train on it instead: +5.25 average across 8 out-of-distribution benchmarks.

Check out our paper and project page.

📚 Paper:arxiv.org/abs/2609.08435re
 🏠 Project Pagtencent-hunyuan.github.io/evolve-scaler/tCp

💬13🔄10❤️163👀17192📊48

Much more AI coming ❤️

Tencent Hy (@TencentHunyuan) — Fri, 11 Sep 2026 02:58:11 GMT

Much more  AI coming ❤️

Ivan Fioravanti ᯅ@ivanfioravanti

This morning I visited Tencent HQ in Beijing!
A marvelous office, from the architecture to the amazing showroom where I've really understood values and services offered by Tencent. 

Thanks to the @TencentHunyuan team in Beijing for the amazing lunch talking about hybrid AI scenarios and more! 

Hy4 Preview is just the tip of the iceberg, there is much more to come! @TencentAI_news Can't wait to see and try what you'll deliver in the near future 🚀

💬8🔄1❤️87👀9783📊14

🚀 AuK is officially here. Nano banana🍌 for audio An open-source foundation model for unified spee...

Tencent Hy (@TencentHunyuan) — Thu, 10 Sep 2026 10:33:40 GMT

🚀 AuK is officially here. Nano banana🍌 for audio

An open-source foundation model for unified speech generation and editing.
Natural-language instructions + reference audio. One interface.
Zero-shot TTS. Instruction-controlled generation. Content editing. Whisper-conversion. De-accent. Timbre/style/emotion edit. Speed/Pitch control. Enhancement, denoising, multi-speaker and music separation.
Also releasing AuK-Flash: 4-step inference. ~4.5× faster under matched conditions.
Code, weights, and demo are live. Try it and share your feedback.

🤗 Paper & upvothuggingface.co/papers/2609.08…JRF
⭐ GitHub & stagithub.com/Tencent-Hunyua…8Yv

Your browser does not support the video tag.

💬45🔄113❤️1018👀94497📊345

✔️Hy4 preview just shipped an upgrade. You flagged it: long thinking + over-verification on complex ...

Tencent Hy (@TencentHunyuan) — Mon, 07 Sep 2026 09:52:10 GMT

✔️Hy4 preview just shipped an upgrade.
You flagged it: long thinking + over-verification on complex tasks.
We optimized it. Now live for everyone.
Same task quality. Fewer turns. Lower in/out tokens.
Bench + human eval both confirm.
We’ll keep iterating fast. Try it and tell us what still breaks.

Try on WorkBuddy: WorkBuddy.ai

💬67🔄77❤️1206👀142023📊258

One more thing about Hy4 preview that is particularly interesting. In AI research. Hy4 preview found...

Tencent Hy (@TencentHunyuan) — Tue, 01 Sep 2026 10:12:32 GMT

One more thing about Hy4 preview that is particularly interesting.
In AI research. Hy4 preview found inference bottlenecks on its own and lifted e2e throughput 31.8% via operator fusion and comms opts — stable across context lengths and concurrency.
MORE ON: hy.tencent.ai

Tencent Hy@TencentHunyuan

🚀 Hy4 preview is here.

770B, 49B active, 1M context.
Built for productivity.
Open source frontier.
Consistent affordable price.
Use it. Tell us what breaks.

More on Hy bloghy.tencent.ai/research/hy4-p…C
HuggingFacehuggingface.co/tencent/Hy4-pr…R
Githubgithub.com/Tencent-Hunyua…L

💬29🔄32❤️460👀36592📊83

sooooo nice

Tencent Hy (@TencentHunyuan) — Mon, 31 Aug 2026 10:57:29 GMT

sooooo nice

苏打白.Dev@sodawhite_dev

我用腾讯的Hy4 preview做了一个《清明上河图》动画效果。

同样的提示词，效果吊打GLM5.3 -flash

Your browser does not support the video tag.

💬10🔄5❤️167👀16423📊27

Everyone can generate a SPIDER-MAN with Hy4 preview

Tencent Hy (@TencentHunyuan) — Mon, 31 Aug 2026 06:02:42 GMT

Everyone can generate a SPIDER-MAN with Hy4 preview

💬8🔄3❤️152👀13775📊28

🤝

Tencent Hy (@TencentHunyuan) — Sat, 29 Aug 2026 16:08:02 GMT

🤝

ModelScope@ModelScope2022

Tencent HY open-sourced Hy4 preview: a flagship MoE with 1M context, and their biggest generational leap measured to date. Apache 2.0.🔥
�modelscope.cn/models/Tencent…AC

⚙️ 770B total params, 49B active per token. Ships with a native MTP layer for speculative decoding out of the box.
📊 Blind eval by 163 internal experts across 203 real engineering tasks: it edges out GLM 5.3 (avg 2.99 vs 2.92) and Kimi K3 (2.99 vs 2.94).
💼 Trained on real production workflows: software engineering, office and data analysis, game development, and scientific research.

🌍 For global usmodelscope.ai/models/Tencent…OFpfV

💬3🔄3❤️101👀10532📊12

Hy4 preview runs in vLLM

Tencent Hy (@TencentHunyuan) — Sat, 29 Aug 2026 16:06:56 GMT

Hy4 preview runs in vLLM

vLLM@vllm_project

@TencentHunyuan's Hy4-preview runs in vLLM from day 0, verified on NVIDIA GPUs. 🎉

- 770B total, 49B active, 256 routed experts plus one shared
- 1M context, but each query attends to just 2048 tokens
- Only 21 of the 78 layers compute their own sparse index, the other 57 reuse one
- A 10B MTP layer ships inside the checkpoint, 0.7B of it active, draft depth 3

Tencent's HPC-Ops attention and MoE kernels have been in vLLM main since Hy3.

VLLM_ENABLE_HPC_OPS=1 vllm serve tencent/Hy4-preview-FP8 -tp 8

Thanks @TencentHunyuan for the preview weights! 🙌

recipes.vllm.ai/tencent/Hy4-pr…yZb

💬1🔄6❤️140👀14763📊15

We compressed Hy4-preview from 1.5TB to ～200GiB GGUF and it still works well ! Meet MIX-STQ1_0.The...

Tencent Hy (@TencentHunyuan) — Sat, 29 Aug 2026 05:31:29 GMT

We compressed Hy4-preview from 1.5TB to ～200GiB GGUF and it still works well ! 

Meet MIX-STQ1_0.The trick isn’t just going low, it’s deciding where: calibration data picks each layer’s bit-width, some down to 1.31-bit STQ1_0, some up to 2.06-bit IQ2_XXS. Same budget, lower error. 

Accuracy barely moves vs BF16
📊 MCP Atlas 83.7→83.2
📊 SWE-Bench multi 82.9→81.3
📊 MRCR 81.3→81.1
📊 IFBench 73.5→72.5

See the details on HF :  AngelSlim/Hy4-preview-GGUF

Weights & low-bit GGUFhuggingface.co/AngelSlim/Hy4-…T9#LLM#Quantizationa#llamacppm#Hyp #Hy

💬84🔄196❤️1838👀632536📊487

Try Hy4 preview in Cline

Tencent Hy (@TencentHunyuan) — Sat, 29 Aug 2026 02:34:36 GMT

Try Hy4 preview in Cline

Cline@cline

Tencent Hy4 Preview leads on SWE-bench Pro. 770B, 49B active, 1M context, and their biggest generational leap measured to date.

It’s exciting to see another open weights model compete against the frontier.

Try in Cline with:
1. npm i -g cline
2. /model
3. Select Hy4 preview

💬5🔄6❤️205👀12915📊24

Hy 4 preview, available in OpenCode Go 👏

Tencent Hy (@TencentHunyuan) — Sat, 29 Aug 2026 02:27:41 GMT

Hy 4 preview, available in OpenCode Go 👏

OpenCode@opencode

Hy4 preview is now available in OpenCode Go

770B/49B · 1M context · built for coding agents

💬8🔄6❤️254👀18557📊31

Free for 2 weeks

Tencent Hy (@TencentHunyuan) — Fri, 28 Aug 2026 13:42:00 GMT

Free for 2 weeks

WorkBuddy@WorkBuddy_AI

Hy4 preview from @TencentHunyuan  is here—and it’s now available on WorkBuddy.

We tested it inside real agent workflows across research, Office tasks, frontend development, and game development.

A few highlights:

→ Strong performance on research tasks that require finding, organizing, and synthesizing information
→ More capable Office workflows, from cross-file analysis to Excel accuracy and PPT visual layout
→ Better frontend execution, with stronger visual quality and interaction design
→ More capable game development, including building playable prototypes from natural-language instructions

What stands out most is how well Hy4 preview fits into end-to-end agent workflows — not just answering questions, but actually completing multi-step productivity tasks.

Hy4 preview is FREE to use on WorkBuddy for the next 2 weeks.

And Hy3 will remain FREE through the end of September.

Try Hy4 preview on WorkBuddy with your own real-world tasks and see how it performs.

💬37🔄41❤️713👀45471📊140

Keep going 💪

Tencent Hy (@TencentHunyuan) — Fri, 28 Aug 2026 08:23:17 GMT

Keep going 💪

Arena.ai@arena

Big news: Hy4 preview by @TencentHunyuan just landed ~#5 in the Code Arena: WebDev with 1633 pts (AutoEval).

This is a significant improvement from Hy3 at #31 overall (+115 pts)! Among open models, Hy4 preview is ~#3, compared to Hy3 at #7.

Note: this is an early AutoEval score, in which a Reward Model trained on Arena's human preference data casts automatic votes in place of live votes. We’ll continue to see how scores converge as more live human votes come in. See thread for more info on the methodology behind AutoEval.

Congrats to the @TencentHunyuan team on this release!

💬9🔄8❤️376👀21095📊44

🚀 Hy4 preview is here. 770B, 49B active, 1M context. Built for productivity. Open source frontier....

Tencent Hy (@TencentHunyuan) — Fri, 28 Aug 2026 06:23:31 GMT

🚀 Hy4 preview is here.

770B, 49B active, 1M context.
Built for productivity.
Open source frontier.
Consistent affordable price.
Use it. Tell us what breaks.

More on Hy bloghy.tencent.ai/research/hy4-p…C
HuggingFacehuggingface.co/tencent/Hy4-pr…R
Githubgithub.com/Tencent-Hunyua…L

💬339🔄560❤️5366👀2845016📊1412

Thank you. We’re working on it — more soon.

Tencent Hy (@TencentHunyuan) — Wed, 26 Aug 2026 06:54:09 GMT

Thank you. We’re working on it — more soon.

Jen Zhu@jenzhuscott

Tencent @TencentHunyuan - 

I’m ready for Hy4.  🚀

💬15🔄17❤️472👀52114📊60

Hy Agent Harness For Video Generation.

Tencent Hy (@TencentHunyuan) — Fri, 21 Aug 2026 12:10:26 GMT

Hy Agent Harness For Video Generation.

Tianyu Pang@TianyuPang1

HyCreator: an agent harness for long video generation

🚀 End2end Auto Mode
⏲️ Generate 10-minute scale films with zero human intervention
👨‍🦰 Switch to real-time interactive editing whenever you want

Apply for early acceshycreator.tencent.comXLh

Your browser does not support the video tag.

💬5🔄20❤️296👀34156📊59

Our translation models are now on @OpenRouterAI 🌐 • Hy-MT2-1.8B • Hy-MT2-30B-A3B Built for real-w...

Tencent Hy (@TencentHunyuan) — Fri, 21 Aug 2026 08:03:54 GMT

Our translation models are now on @OpenRouterAI 🌐

• Hy-MT2-1.8B
• Hy-MT2-30B-A3B

Built for real-world translation across 33 languages, with strong multilingual instruction following for structured, contextual, and style-guided tasks.

Let’s try them out ↓openrouter.ai/tencentD

💬17🔄25❤️400👀25209📊90

8x the normal usage on OpenCode👌

Tencent Hy (@TencentHunyuan) — Fri, 21 Aug 2026 02:54:16 GMT

8x the normal usage on OpenCode👌

OpenCode@opencode

Hy3 will get 8x the normal usage until August 30th

that's $480 of usage for $10

available to all OpenCode Go subscribers

💬13🔄1❤️365👀21723📊43

Hyra keeps pushing the frontier. Since our last math update, four more advances have landed: 📐 3D B...

Tencent Hy (@TencentHunyuan) — Wed, 19 Aug 2026 09:06:37 GMT

Hyra keeps pushing the frontier.
Since our last math update, four more advances have landed:
📐 3D Blaschke–Lebesgue: raised the universal lower bound for constant-width bodies from 0.380799w³ → 0.411040w³ (now 97.9% of the conjectured Meissner optimum).
∫ Beurling–Ahlfors: improved the best uniform (L^p) coefficient from 1.575 → 1.523958, moving closer to Iwaniec’s conjectured sharp bound.
± Partial Hadamard matrices: extended asymptotic counting from the cubic scale down to the near-quadratic regime.
[,] Operator commutators: improved the cost of approximating the identity from Tao’s O(log⁵(1/ε)) to O(log³), narrowing the gap toward the known logarithmic lower bound.
We are making open-source models better at solving open problems!
Check the results:github.com/Tencent-Hunyua…U

💬5🔄8❤️147👀17899📊28

📄New Research on Self-Evolving Agents: When AI agents modify themselves, how do we know they actual...

Tencent Hy (@TencentHunyuan) — Wed, 12 Aug 2026 07:42:34 GMT

📄New Research on Self-Evolving Agents:
When AI agents modify themselves, how do we know they actually got better?

We present Diving into Reliable Self-Evolving Agents: A Survey—a systematic map of how agents self-evolve and what evidence is needed to trust each update.
The survey:
🔹 Defines an L0–L4 taxonomy for self-evolving agents
🔹 Introduces a reliability ladder for trustworthy updates
🔹 Curates 549 works in an open companion catalog
One core principle: no update should control the only evidence used to accept itself.

Explore the full survey ↓
📄 Papopenreview.net/forum?id=CGO1h…dF2FP
🌐 Prowkqdzkd.github.io/Awesome-Reliab…9EIPf0
💻 Ggithub.com/wkqdzkd/Awesom…now4J9X

💬14🔄50❤️379👀24604📊120

Introducing Hy3D WorldClaw——an agentic workflow that generates large scale 3D open worlds from text...

Tencent Hy (@TencentHunyuan) — Tue, 11 Aug 2026 06:48:22 GMT

Introducing  Hy3D WorldClaw——an agentic workflow that generates large scale 3D open worlds from text prompts.🚀🚀🚀
Not video, Not Gaussian Splatting, Every scene generated by WorldClaw is freely explorable and built entirely from editable, game-ready 3D assets with high-quality geometry and textures.
Project Pagtencent-hunyuan.github.io/Hunyuan3D-Worl…b1S

Your browser does not support the video tag.

💬80🔄262❤️2502👀233235📊832

3⃣3⃣3⃣

Tencent Hy (@TencentHunyuan) — Thu, 06 Aug 2026 07:36:27 GMT

3⃣3⃣3⃣

Tencent 腾讯@TencentGlobal

Built for practical AI, Tencent Hy3 combines flagship-level performance with a compact, cost-efficient design. Bringing advanced reasoning and agent capabilities to research, coding, office productivity, design and cloud workflows.

Now available globally across Tencent products, APIs and developer platforms.

Try Hy3 today.
Hy3 Official Website: hy.tencent.ai
Hy3 API Pricing: openrouter.ai/tencent/hy3
Tencent WorkBuddy: workbuddy.ai
Tencent Design Miora: miora.design
Tencent Cloud TokenHub: tencentcloud.com/products/token…

Your browser does not support the video tag.

💬5🔄3❤️124👀17284📊19

🚀 Hy3 is now live in WorkBuddy global—free for users worldwide through August 31, 2026 (PT). Resea...

Tencent Hy (@TencentHunyuan) — Wed, 05 Aug 2026 02:59:51 GMT

🚀 Hy3 is now live in WorkBuddy global—free for users worldwide through August 31, 2026 (PT).

Research, analyze data, create documents and presentations, and handle workplace communications in an agentic AI workspace—no setup required.

Hy3 is also available through Tencent Design Miora and Tencent Cloud TokenHub.

Try it:workbuddy.air
#TencentHyy#Hy33#WorkBuddyy#Mioraa#TencentCloudd

💬32🔄57❤️399👀154063📊134

AI that finally hears you. 👂 Hy ASR 3.0 preview drops today. Powered by Hy3’s language brain → ...

Tencent Hy (@TencentHunyuan) — Tue, 04 Aug 2026 09:58:55 GMT

AI that finally hears you. 👂

Hy ASR 3.0 preview drops today. 

Powered by Hy3’s language brain 
→ precision ASR + deep semantic understanding

Wins on:
• general recognition
• context awareness
• multi-scenario robustness
• dialect coverage

Messy real-world audio? Clean, coherent transcripts that nail your intent. 

Open + internal benches: leading.
Multilingual WER ≈ 3%. 

Voice just got an upgrade. 🔥#HyASRS#TencentAIA#Hy3y3

💬23🔄28❤️455👀29859📊120

Hy-MT2 keeps gaining momentum. Since its open-source release in May: → 700K+ downloads 🌟 → Hy-MT2-...

Tencent Hy (@TencentHunyuan) — Fri, 31 Jul 2026 08:32:29 GMT

Hy-MT2 keeps gaining momentum.

Since its open-source release in May:
→ 700K+ downloads 🌟
→ Hy-MT2-1.8B reached #1 on the Hugging Face trending, with 30B-A3B reaching #4 🥇
→ 70+ verified product and project integrations 💻
→ Broader Hy-MT ecosystem support across Apple MLX-LM, Microsoft ONNX Runtime, NVIDIA NeMo, LLaMA-Factory, and more 👯
→ Real-world adoption, including real-time multilingual translation of livestream comments on Bilibili 📺

And now, Hy-MT2-30B-A3B is officially available in GGUF format—addressing one of the community’s most-requested deployment needs and making local inference easier.

Ready to run Hy-MT2-30B-A3B locally? Try the new GGUF relehuggingface.co/tencent/Hy-MT2…TVV7I
Explore Hy-MT2:
Hugginghuggingface.co/collections/te…McENW
Modelsmodelscope.cn/collections/Te…Yk9zQ
Gigithub.com/Tencent-Hunyua…Ac#TencentHye#HyMT2H#OpenSourceource

💬7🔄21❤️261👀18551📊53

join us,make different🥳

Tencent Hy (@TencentHunyuan) — Fri, 31 Jul 2026 02:01:37 GMT

join us,make different🥳

Shunyu Yao@ShunyuYao12

Hy ai4s is hiring :)

💬5🔄0❤️150👀30604📊19

For a finite set of integers (A), how much faster can (|A+A|) grow than (|A-A|)? A 1969 theorem gav...

Tencent Hy (@TencentHunyuan) — Thu, 30 Jul 2026 02:33:16 GMT

For a finite set of integers (A), how much faster can (|A+A|) grow than (|A-A|)?

A 1969 theorem gave an upper bound of 2 for the exponent. For more than 50 years, the best constructions barely exceeded 1.1.

With help from our research agent Hyra and the Hy3 model, we found an explicit construction showing that the optimal exponent is exactly 2.

A 50-year-old problem, solved.

Paper: arxiv.org/abs/2607.27199
Hyra blog: hy.tencent.ai/research/hyra
Formal proof: github.com/linhaowei1/sum…

💬39🔄192❤️1586👀394688📊372

🚀 We’ve open-sourced AngelSpec, an end-to-end speculative decoding framework supporting both train...

Tencent Hy (@TencentHunyuan) — Wed, 29 Jul 2026 12:43:55 GMT

🚀 We’ve  open-sourced AngelSpec, an end-to-end speculative decoding framework supporting both training and deployment.

On Hy3-A21B, DFly delivers a 1.98–2.40× end-to-end speedup over autoregressive decoding across tested concurrency levels from 4 to 64, with 10.5–11.8% higher throughput than DFlash.

Training code and Hy3-A21B MTP/DFly drafter weights are now available:
GitHub:github.com/Tencent/AngelS…Y
Paper:arxiv.org/abs/2607.25852v
Docs:angelspec.readthedocs.ioR
Hugging Face:huggingface.co/collections/An…C
ModelScope:modelscope.cn/collections/An…Y#Hy33#AngelSpecc#OpenSourcee

💬19🔄101❤️658👀43965📊182

Test Hy3 in GO. 💗

Tencent Hy (@TencentHunyuan) — Thu, 23 Jul 2026 03:05:26 GMT

Test Hy3 in GO.  💗

OpenCode@opencode

Hy3 now available in Go

text · 256K context

latest model from Tencent

💬16🔄5❤️440👀28639📊54

Agent & Coding 🔥🔥🔥

Tencent Hy (@TencentHunyuan) — Wed, 22 Jul 2026 08:29:23 GMT

Agent & Coding 🔥🔥🔥

Arena.ai@arena

Hy3 by Tencent is #5 in Agent Arena for open-weight models (#25 overall)! It also ranks as the #2 open model in the Frontend Code Arena (#16 overall)!

In Agent Arena: Hy3 lands at #25 overall (net -2.2%). Hy3 has strengths in tool-use (recovering well from CLI/bash errors, +2.6% and #25), its biggest weakness is steerability as it struggles to course-correct when users push back, coming in at -7.1% (#30).

Agent Arena measures models on millions of real-world, long-horizon agentic tasks. Models get web search, filesystem, and terminal tools to complete complex workflows: writing code, creating slide decks, researching the web, building apps, and analyzing documents.

We use causal tracing methodology to measure a model's net improvement, which indicates how much it improves outcomes relative to the average model.

Below we break down how Hy3 scored across 5 key signals, drawn from tasks submitted by a global community of users. Here’s an overview on the signals:

User-satisfaction proxies
- Confirmed Success: an explicit "yes that worked" feedback from the user
- Praise vs. Complaint: implicit sentiment in users reactions
- Steerability: can the model course-correct when you push back?

Tool-use proxies
- Bash Recovery: how it recovers from CLI errors (primary signal for tool use)
- Tool Hallucination: does it call tools that don't exist

Congrats to the @TencentHunyuan team on this release!