AI at Alibaba International

As the #AI Business team of Alibaba International Digital Commerce Group (AIDC) , we work on fundation modles to revolutionize digital commerce

[AI at Alibaba International @AI_AlibabaInt]

Congrats to our team! 🎉5 papers accepted by #ICLR2026 Check out our research lineup below!
@iclr_conf

[AI at Alibaba International @AI_AlibabaInt]

Getting Your LLMs Ready for Reinforcement Learning with Lightweight SFT
🔍https://t.co/uYazPWkMJu
#ICLR2026 #LLM #SFT

[AI at Alibaba International @AI_AlibabaInt]

Congrats to our team! 🎉5 papers accepted by #ICLR2026 Check out our research lineup below!
@iclr_conf

[AI at Alibaba International @AI_AlibabaInt]

Adaptive Debiasing Tsallis Entropy for Test-Time Adaptation
🔍 https://t.co/Sx0pfromRR

#ICLR26

[Longyue Wang @wangly0229]

🌍Our @AI_AlibabaInt has 10 papers accepted by #ICLR2026 and #ICASSP2026!🔥

[ModelScope @ModelScope2022]

🚀 Ovis-Image (7B) is live on ModelScope!

✅Delivers frontier-level text rendering—on par with 20B-class models like Qwen-Image and even competitive with GPT-4o on text-heavy tasks.
✅Sharp, layout-aware output for posters, banners, logos, UI mocks, and infographics.
✅Runs fast and lean—deployable on a single high-end GPU.

Small model. Big text fidelity.
👉 https://t.co/lJPuaWjMr2

[AI at Alibaba International @AI_AlibabaInt]

Github:
https://t.co/j1ld3xuyxH

[AI at Alibaba International @AI_AlibabaInt]

Try our latest work Ovis Image 7B 🎉

[AK @_akhaliq]

Ovis-Image is out on Hugging Face

a 7B text-to-image model specifically optimized for high-quality text rendering

[AI at Alibaba International @AI_AlibabaInt]

Model：
https://t.co/oOXpQNjW9G

[Chenyang Lyu @Chenyang_Lyu]

We released Marco-Voice months ago,now the pre-trained model weights and an online demo for Marco-Voice are LIVE!

Try the Demo: https://t.co/yTBpDCS1KW
Find the Project on GitHub: https://t.co/b3CLrkCgIn
Download the Models on Hugging Face: https://t.co/TESm6i19Rv

[Pixelle AI @Pixelle_AI]

☕️A coffee takes 5 minutes.  A full video now takes even less.

Introducing Pixelle-Video — ⚡️AI Fully Automated Short Video Engine.
Built fully on ComfyUI workflows & backend.

Input an idea → get narration, images, layout, TTS, all in one pipeline.Try Pixelle-Video and share your outputs with us !

👉GitHub: https://t.co/LeUHqje8zp

[Longyue Wang @wangly0229]

🎯Introducing Marco Search Agent — an open-source project Towards Real-world & Challenging Agentic Search! @AI_AlibabaInt

First of all, we release two agent benchmarks:
🔥 DeepWideSearch — Benchmarking Search Agents on Depth and Width in Information-Seeking
🔥 HSCodeComp — Benchmarking Search Agents in Hierarchical Rule Application (Harmonized System Codes Prediction as a testbed)

More Details:
🌟GitHub: https://t.co/dQ9MCNQuZc
📕DeepWideSearch Paper: https://t.co/i9LIWCRzFY
📗HSCodeComp Paper: https://t.co/cLlTnaK1Ur

[AI at Alibaba International @AI_AlibabaInt]

Thrilled to share our breakthrough at #WMT2025!
Our Marco-MT (Team Algharb) achieved something remarkable: we ranked #1 in English→Chinese translation, which not only outperforms leading AI systems like Claude-4 and GPT-4.1, but also surpasses human translators, proving our model's excellence in general translation tasks.

Among 13 language pairs we competed in, Maroc-MT-Algharb achieves:
🏅6 First Places
🥈4 Second Places
🥉2 Third Places

We did this with key innovations:
• Novel M2PO translation paradigm
• Two-stage SFT + CPO+MPO reinforcement learning
• Hybrid decoding with word alignment & MBR

Learn more:
🔧Demo: https://t.co/dKdudmepoB
📄Technical Report: https://t.co/K6b6PX9MhR
🤗Hugging Face: https://t.co/lnBXqhsjoa

#NLP #WMT2025 #MarcoMT

[AI at Alibaba International @AI_AlibabaInt]

Wan animate is available on Pixelle MCP🎉
Try to create your kongfu cat video here:
https://t.co/uNSvXoBwGz
Prompt: Generate a motion transfer video based on this video and cat image using wan animate
#AlibabaWan

[Pixelle AI @Pixelle_AI]

🎬 Tried the new wan2.2 animate @Alibaba_Wan and ended up with a Taoist master training alongside his cat 🐱💃
Not sure who’s teaching who, but the sync is wild 😂Cooked this up in Pixelle-MCP ⚡ — our own open-source project (yep, also from the Alibaba family 💡).Testing new models has never been this fun.
👉 Who else is playing with wan2.2 animate? Show me your craziest outputs!
#AI #PixelleMCP #opensource

[AI at Alibaba International @AI_AlibabaInt]

Thanks @slatornews for featuring Marco-Voice! 🗣️
Pushing boundaries in TTS with unified voice cloning & emotion control.
Check out our work and join us in advancing expressive speech synthesis!
https://t.co/xTfwBQ8pCJ
#AI #SpeechTech #MarcoVoice

[Slator @slatornews]

Alibaba unveils Marco-Voice, a new text-to-speech system that combines #voicecloning 🗣️ and emotional #speechsynthesis, 😐😄😠😢😮 delivering more natural and expressive synthetic #speech in Mandarin and English.
@Chenyang_Lyu @wangly0229 @AlibabaGroup
https://t.co/Q2wXtU2dez