Pinned Tweet
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.
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.
32
41
536
48,707
🚀 Introducing Hy-Nomas: An Open Source Free-Form Interactive Video World Model
Key Features:
* Free-form interaction with text prompts and camera trajectories
* Long-horizon generation with world memory
* Lightweight LoRA adaptation for specialized domains such as robotics
* on Consumer GPUs
Test it and drop feedback anytime！
🔗 Website: hy-nomas.hy.tencent.com
💻 GitHub:github.com/Tencent-Hunyua…
🤗 Hugging Face: huggingface.co/tencent/Hy-Nom…
🎥 Demo: hy-nomas.hy.tencent.com/demo
#Hy#WorldModels #VideoGeneration #GenerativeAI
Key Features:
* Free-form interaction with text prompts and camera trajectories
* Long-horizon generation with world memory
* Lightweight LoRA adaptation for specialized domains such as robotics
* on Consumer GPUs
Test it and drop feedback anytime！
🔗 Website: hy-nomas.hy.tencent.com
💻 GitHub:github.com/Tencent-Hunyua…
🤗 Hugging Face: huggingface.co/tencent/Hy-Nom…
🎥 Demo: hy-nomas.hy.tencent.com/demo
#Hy#WorldModels #VideoGeneration #GenerativeAI
35
119
896
59,458
📢 Introducing WorkForge: a scalable framework that constructs verifiable training environments for long-horizon work agents from real-world resources.
Starting from expert workflows, it builds authentic workspaces, extracts checkable facts, and derives tasks plus dual c) verifiers grounded in observable evidence.
🚀 We scale to 16.7K environments across 40 professional domains covering 60 file types, enabling consistent gains on long-horizon work benchmarks as data volume and interaction horizon increase.
📚Paper: arxiv.org/pdf/2610.04906 @zhang_leo54389 @ZoeyyyZhououou @Be1ong1
#Tencent Hy#AIAgents #WorkAgents #AgentTraining
Starting from expert workflows, it builds authentic workspaces, extracts checkable facts, and derives tasks plus dual c) verifiers grounded in observable evidence.
🚀 We scale to 16.7K environments across 40 professional domains covering 60 file types, enabling consistent gains on long-horizon work benchmarks as data volume and interaction horizon increase.
📚Paper: arxiv.org/pdf/2610.04906 @zhang_leo54389 @ZoeyyyZhououou @Be1ong1
#Tencent Hy#AIAgents #WorkAgents #AgentTraining
7
17
162
13,972
🚀 Hy Memory 2.0 is out — open-source model, framework, and MemHub.
Your private memory steward.
Framework handles chat, work & code memories.
#1 open-source on 4 benchmarks.
Hy-Memory-30B-A3B beats Opus 5 across major frameworks.
MemHub (CLI + Mac) syncs across 10+ AI tools.
💻➡️://t.co/uP58VJanj8
🔗 Model: huggingface.co/tencent/Hy-Mem… #HyMemory #AgentMemory
Your private memory steward.
Framework handles chat, work & code memories.
#1 open-source on 4 benchmarks.
Hy-Memory-30B-A3B beats Opus 5 across major frameworks.
MemHub (CLI + Mac) syncs across 10+ AI tools.
💻➡️://t.co/uP58VJanj8
🔗 Model: huggingface.co/tencent/Hy-Mem… #HyMemory #AgentMemory
65
125
1,315
155,599
AMAZING😍
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 👇
1
2
48
6,129
more 🔛 github.com/Tencent-Hunyua…
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 ed 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.
📄 Paper: arxiv.org/abs/2609.30199
🌐 Website & leaderboard: explorationbench.com
📝 Blog: explorationbench.com/blog/
💻 Code (coming soon): github.com/Tencent-Hunyua…
0
1
45
7,031
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 ed 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.
📄 Paper: arxiv.org/abs/2609.30199
🌐 Website & leaderboard: explorationbench.com
📝 Blog: explorationbench.com/blog/
💻 Code (coming soon): github.com/Tencent-Hunyua…
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 ed 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.
📄 Paper: arxiv.org/abs/2609.30199
🌐 Website & leaderboard: explorationbench.com
📝 Blog: explorationbench.com/blog/
💻 Code (coming soon): github.com/Tencent-Hunyua…
14
45
428
70,902
3 tracks. 3 winners. Go and Create. 🥇
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. 👇
1
2
54
8,254
live on 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
8,144
🚀 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 avaico/16zTeVLb1s
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 avaico/16zTeVLb1s
34
60
645
51,593
⚡️ 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/100116
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/100116
10
8
194
15,071
Tencent Hy retweeted
Tencent dropped Hy Image3.5 preview!
Great improvement and nice quality. Much better than HY Image 3.0 Instruct.
same approach: deep reasoning before editing/creating.
- doesn’t need detailed prompts
- follows instructions correctly
- solid composition understanding
Pushes o a new level.
Yeah, results aren’t always perfect, but it doesn’t destroy faces or human features.
Great improvement and nice quality. Much better than HY Image 3.0 Instruct.
same approach: deep reasoning before editing/creating.
- doesn’t need detailed prompts
- follows instructions correctly
- solid composition understanding
Pushes o a new level.
Yeah, results aren’t always perfect, but it doesn’t destroy faces or human features.
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…
Priced for everyone. $0.024 per image on Tencent Cloud API.
We ogenerate — your reference images are free.
Two weeks free (Only in OnSolo and Miora):
Miora miora.design
OnSolo onsolo.ai
Try it and tell us where it breaks.
3
9
93
8,920
ComfyUI ✖️ Hy Image3.5 preview
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, and illustration styles
→ Identity and product features that hold through scene, outfit, and style changes
2
3
85
7,929