China AI Native Industry Insights – 20250414 – Kunlun Tech | ByteDance | Rokid | more
Explore Kunlun Tech’s open-source launch of Skywork-OR1 models, ByteDance’s pioneering Multi-SWE-bench for multi-language code repair, and Rokid Glasses’ AI-driven teleprompter innovation. Discover more in Today’s China AI Native Industry Insights.
1. Kunlun Tech open-sources Skywork-OR1: Leading 7B and 32B reasoning models
🔑 Key Details:
– Kunlun Tech releases Skywork-OR1 series, featuring 7B and 32B models for math and code reasoning
– Outperforms Alibaba QwQ-32B and rivals DeepSeek-R1 in AIME and LiveCodeBench benchmarks
– Three models released: Math-7B, 7B-Preview, and 32B-Preview, all with full training code, weights, and datasets
– Introduces avg@k metric for more stable and reliable evaluation of reasoning ability
– Fully open-sourced on GitHub and Hugging Face, with detailed technical blog on Notion
💡 How It Helps:
– Developers: Access powerful reasoning models with full reproducibility
– Researchers: Study transparent training pipeline and GRPO-based reinforcement learning
– AI community: Gains a strong open-source alternative for logic, math, and code tasks
🌟 Why It Matters:
Skywork-OR1 sets a new standard for open-source reasoning models, showing that high performance doesn’t require massive model size. Kunlun Tech’s strategy supports accessible, reproducible AI research and pushes forward the global reasoning model ecosystem.
Original Chinese article: https://mp.weixin.qq.com/s/tEIaqE07Z5j_tAl31ElK8w
English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https%3A%2F%2Fmp.weixin.qq.com%2Fs%2FtEIaqE07Z5j_tAl31ElK8w
Video Credit: The original article
2. ByteDance’s Multi-SWE-bench: The First Multi-Language Code Repair Benchmark Open-Sourced
🔑 Key Details:
– ByteDance Doubao Big Model team introduces Multi-SWE-bench, the first multi-language benchmark for evaluating AI code repair capabilities
– Covers 7 programming languages: Java, TypeScript, C, C++, Go, Rust, and JavaScript
– 1,632 real-world repair tasks sourced from GitHub with difficulty grading (Easy, Medium, Hard)
– Aims to enhance AI’s cross-language generalization and real problem-solving abilities
– Multi-SWE-RL also released for reinforcement learning in code environments
– Fully open-source on Hugging Face and GitHub
💡 How It Helps:
– Developers: Provides standardized, reproducible benchmark for multi-language code repair
– Researchers: Expands AI’s capability in solving complex programming tasks across languages
– AI Community: Fosters collaboration for dataset expansion and RL training in real-world software engineering
🌟 Why It Matters:
Multi-SWE-bench bridges gaps in existing benchmarks, offering a comprehensive test for code generation models across multiple languages. It pushes AI towards becoming a true general-purpose software agent, driving forward the evolution of intelligent code repair.
Original Chinese article: https://mp.weixin.qq.com/s/hoFAPJc4WCeWMBiFSZ1mkg
English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https%3A%2F%2Fmp.weixin.qq.com%2Fs%2FhoFAPJc4WCeWMBiFSZ1mkg
Video Credit: The original article
3. Rokid Glasses Introduces AI-driven Teleprompter with Speech Collaboration
🔑 Key Details:
– Rokid Glasses introduces an AI-powered teleprompter that automatically adjusts manuscript speed based on the speaker’s rhythm
– New patent includes three key innovations:
– Multimodal speech recognition for real-time speech analysis and dialect support
– Dynamic speech speed adaptation that tracks speech pace and adapts the scroll
– Multi-scenario speech matching for accurate manuscript positioning during improvisation or skipping
– Eliminates reliance on manual controls, reducing distractions and psychological pressure for speakers
– Aims to make AI an active collaborator, enhancing speaker focus on content delivery
💡 How It Helps:
– Speakers: Focus 90% on content and audience, reducing cognitive load
– Event Organizers: Seamlessly integrates AI with live speeches, improving speech flow
– Developers: Offers a new way of human-AI collaboration with intelligent context recognition
🌟 Why It Matters:
This innovation transforms the teleprompter from a simple tool to an intelligent partner, enabling more natural, confident speeches and paving the way for future human-machine collaboration in dynamic environments.
Original Chinese article: https://mp.weixin.qq.com/s/5QHgjNMBF9KRmiV9RkvQIQ
English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https%3A%2F%2Fmp.weixin.qq.com%2Fs%2F5QHgjNMBF9KRmiV9RkvQIQ
Video Credit: The original article
That’s all for today’s China AI Native Industry Insights. Join us at AI Native Foundation Membership Dashboard for the latest insights on AI Native, or follow our linkedin account at AI Native Foundation and our twitter account at AINativeF.