When ancient movements meet modern algorithms, AI carries on the legacy of martial arts.
Powered by Baidu ERNIE, the martial arts large model can capture movements, preserving and recording them in digital form.
This approach allows the traditional oral transmission and master-apprentice lineage to make a digital leap.
What if martial arts could be learned from AI? ๐ค
Weโve built an AI model for Chinese Wushu โ turning master-to-student teachings into visual, easy-to-learn experiences.
Making tradition easier to learn, and easier to pass on. โจ
Zack Williams brought one of the world's oldest writing systems to the ERNIE AI Developer Challenge: ancient cuneiform tablets.
Using PaddleOCR, he built NabuOCR to help read cuneiform from tablet images.
See the story behind @BoatbomberRBLX's winning project ๐
As AI agents take on more work, it's worth asking what we should measure.
Tokens tell you what you spent.
DAA, or Daily Active Agents, tells you what you got back ๐
Here is a quick addition to your metrics vocabulary: DAA.
Short for Daily Active Agents, it is the agent era's equivalent of DAU. Where tokenomics tracks cost, DAA tracks output โ how much work agents are actually getting done.
See the full comparison โ
Great to see ERNIE 5.1 recognized among todayโs leading models in real-world evaluations.
Weโll keep building practical, powerful AI for users and developers. Step by step. ๐
With its new mobile app, our AI-native agent DuMate is becoming a more capable AI assistant for everyday work across devices.
Designed to execute complex, multi-step workflows, it syncs tasks between mobile and PC in real time, while drawing on Baidu Search AI API, Miaoda, Famou Agent, Baidu Baike, and other core capabilities.
Since launch, DuMate has expanded into more scenarios, and achieved SOTA-level results on multiple agent benchmarks, including PinchBench and DeepResearch Bench.
Robin proposed Daily Active Agents (DAA) as a defining metric for the agent era, a counterpart to DAU in the mobile internet era.
While token consumption reflects cost more than value, DAA brings the conversation back to output.
As Robin noted, to measure the health of a platform or ecosystem, more attention should be paid to the DAA metric โ the number of agents actively working and delivering results.
ERNIE 5.1 is here ๐
ERNIE 5.1 significantly reduces pretraining cost while compressing total parameters to ~1/3 and activated parameters to ~1/2 โ using only ~6% of the pretraining cost compared to models at similar scale, while achieving leading performance in its class.
๐กKey highlights:
1/ Strong agentic performance approaching leading frontier models. ERNIE 5.1 surpasses DeepSeek-V4-Pro on both ฯ3-bench and SpreadsheetBench-Verified.
2/ Strong world knowledge and creative writing capabilities, with GPQA and MMLU-Pro performance approaching leading closed-source models, and creative writing ability nearing Gemini 3.1 Pro.
3/ Frontier-level reasoning performance. ERNIE 5.1 scores 99.6 on the challenging AIME26 benchmark with tools, second only to Gemini 3.1 Pro.
4/ Deep search capability. On May 9, ERNIE 5.1 ranked #4 globally and #1 among Chinese models on the Arena Search leaderboard with a score of 1223.
ERNIE 5.1 is now available on ERNIE and the Baidu AI Studio Model Playground:
๐https://t.co/qhd67Lg3B4
๐https://t.co/AaQSqDmVGU
๐https://t.co/uCNiypIu1q
Baidu Create 2026 is coming up fast, and the agenda is packed!
Beyond the main forum, our flagship developer conference returns with special forums on AI infrastructure, agent development, real-world applications, and more โ plus plenty to explore on site.
Take a look ๐
ERNIE 5.1 Preview just went live ๐ With a lighter, more efficient architecture, it delivers strong performance at its scale. And this is just the start โ more ERNIE model updates to come at Baidu Create 2026.
Introducing ERNIE 5.1 Preview โ now live! ๐
Ranked #13 globally and #1 among Chinese labs on @arena 's Text Arena.
Top-10 worldwide across:
๐ Math โ #9
โ๏ธ Legal & Government โ #1
๐ผ Business, Management & Financial Ops โ #4
๐ป Software & IT Services โ #7
Built on the strong pre-training foundation of ERNIE 5.0, ERNIE 5.1 Preview compresses total parameters to ~1/3 and activated parameters to ~1/2 of its predecessor, while using only ~6% of the pre-training cost of comparable models โ delivering leading foundational performance at its scale.
Try it now ๐ https://t.co/jNwEbwrKps
More new models are on the way โ stay tuned ๐
GenFlow 4.0 is here ๐From parallel Office Agents to a fully integrated AI workspace, weโre pushing what AI agents can do at scale.
More to come at Baidu Create 2026 ๐
GenFlow 4.0 is live, and it's already serving 100M+ monthly active users with 200M tasks completed each month! ๐
Jointly released by Baidu Wenku and Baidu Drive, GenFlow 4.0 is a major upgrade to our general AI Agent, with a fully revamped Office Agent at its core.
Users can now invoke PowerPoint, Excel, and Word Agents in parallel from a single prompt.
GenFlow 4.0 is also deeply integrated with OpenClaw, deployable in one click from the Baidu Drive PC or mobile app, turning Baidu Drive into a personal AI workspace.
More to come at Baidu Create 2026 in Beijing May 13-14, where we'll explore this year's theme: "Agents at Scale."
AI meets agriculture ๐พ
Weโve launched our first AI agent for farmingโco-developed with leading agricultural scientists.
Farmers can now get expert advice through simple conversations and work more efficiently. ๐๐ก
Introducing ERNIE-Image โ Our Open 8B Text-to-Image Model is Live!๐
> Only 8B DiT params, #1 open weights model on GenEval, OneIG, & LongTextBench
> Precise text rendering in English, Chinese, and more
> Complex instruction following & multi-object control
> Posters, manga, multi-panel layouts with structural coherence
> Broad styles: photography, cinematic tones, graphic design, and more
> Easy to deploy: Runs on 24GB VRAM
Start creating with ERNIE-Image ๐
๐ค Model:
> ERNIE-Image https://t.co/nyooKMlEtk
> ERNIE-Image-Turbo https://t.co/HJz4JvjpSE
๐ Blog: https://t.co/rXGMIVJ401
๐ฅ๏ธ Demo: https://t.co/oiBsdnVfsS
๐ง Github: https://t.co/7GcCmebaYW
๐พ Discord: https://t.co/jZvsNhOb83
Huge congrats on the launch of ERNIE-Image โ a major leap forward!
Weโre excited to see ERNIE-Image pushing the boundary of open text-to-image models with strong performance, efficient design, and real usability.
Welcome everyone to try it out โค๏ธ
โจERNIE-Image is here โ Baidu's open 8B text-to-image model that punches way above its weight.
โ Only 8B DiT params, #1 open weights model on GenEval, OneIG, & LongTextBench
โ Precise text rendering in English, Chinese, and more
โ Complex instruction following & multi-object control
โ Posters, manga, multi-panel layouts with structural coherence
โ Broad styles: photography, cinematic tones, graphic design, and more
โ Easy to deploy: Runs on 24GB VRAM
Two versions:
ERNIE-Image โ SFT model, stronger general quality in 50 inference steps
ERNIE-Image-Turbo โ Optimized for speed and aesthetics in just 8 steps
๐ค Model:
https://t.co/QCZkSFZEIz
https://t.co/Eutm0gurmp
๐ Blog: https://t.co/Gb5uQX4mAU
๐ฅ๏ธ Demo: https://t.co/vikUfdp6KO
๐ง Github: https://t.co/SgFNyxzZKZ
๐พ Discord: https://t.co/05DviJtHU0
Join our Discord for free bot access, weekly challenges, and prizes!
The results for #ACL2026 are in โ Baidu has 23 papers accepted. ๐
Check out some of our accepted papers below๐
1. AttnPO: Attention-Guided Process Supervision for Efficient Reasoning
๐ https://t.co/uamhv83xco
2. RRAtention: Dynamic Block Sparse Attention via Per-Head Round-Robin Shifts for Long-Context Inference
๐ https://t.co/89iW7zyDUq
3. Agentic-R: Learning to Retrieve for Agentic Search
๐ https://t.co/alIG6AJ7NP
4. CoVerRL: Breaking the Consensus Trap in Label-Free Reasoning via Generator-Verifier Co-Evolution
๐ https://t.co/FCuzDdicKv
5. Safety-Utility Conflicts Are Not Global: Surgical Alignment via Head-Level Diagnosis
๐ https://t.co/MNd0NTKj1Y
As one of the most prestigious conferences in computational linguistics and NLP, ACL is known for its rigorous review process and high selectivity. This recognition highlights our ongoing commitment to AI research.โจ
Today we unveiled the top 10 frontier technology inventions of 2025, including breakthroughs spanning large models, deep learning frameworks, AI compute, agents, AI search, digital humans, and autonomous driving.
Behind the accelerating emergence of AI applications lies our deep strength in foundational innovation โ ranking first in China in AI patents for seven consecutive years and ranking among the global leaders in generative AI, deep learning patents, and autonomous driving technology.
See these many of these innovations in action at Baidu World 2025 on November 13!
Less than a week to go! โฐ
Join us live on Nov 13 at 9:30 am UTC+8 / 5:30 pm PT for Baidu World 2025 โ streaming straight from Beijing. Discover our newest AI innovations and see how they're shaping everyday possibilities.
Watch the livestream on X or YouTube!
X: https://t.co/MVVxsDPfGQ
YouTube: https://t.co/YmuBdHV2Id
Qianfan-VL, Baidu AI Cloud's vision-language model series, is now open source! Designed for enterprise-level applications, these multimodal models combine robust general capabilities with advanced performance in OCR and math problem-solving.
Key features:
> Three model sizes (3B, 8B, 70B) with 32K context length for diverse needs
> Chain-of-thought reasoning in 8B/70B for strong performance in chart understanding, math, and visual logic
> Four-stage progressive training pipeline for improved cross-modal alignment and domain enhancement
> High-precision data synthesis pipeline across documents, math, charts, tables, formulas, and OCR tasks
Discover more about Qianfan-VL โ