Recent AI Industry Highlights: New Models, Open Source Releases, and User Milestones

The AI industry continues to evolve at a rapid pace, with recent months seeing a flurry of new model releases, open-source initiatives, and user growth milestones. Zhihu, a leading Chinese knowledge-sharing platform, recently published its "Annual AI Product List" for 2025, offering a comprehensive review of the year's key developments. Curated by global developers and users, the list features top AI products like Doubao, DeepSeek, and Gemi, highlighting their impact on the industry.

Alibaba's Tongyi Qianwen (Qwen) series has been particularly active. On April 2nd, Alibaba launched the Qwen6-Plus model, now accessible via the Alibaba Cloud Bailian API with significant improvements over its predecessor. The series also expanded with Qwen-Image, a 2-billion-parameter multimodal diffusion transformer (MMDiT) for image generation, and Qwen3-Coder, an open-source "world-facing intelligent programming engine". Additionally, the Qwen team introduced a "Deep Research" feature, designed to help users plan and execute complex research tasks step-by-step. As AI integrates deeper into various industries, Alibaba Cloud launched the Double 11 "AI Smart Benefits Season" to provide enterprises and developers with favorable policies and benefits for its full-stack AI cloud services. However, the platform faced a temporary outage recently, with the "Alibaba Qianwen down" topic trending on Weibo—prompting an immediate response from the official team.

DeepSeek, another key player, made waves with its V4Pro model release (launched alongside Musk's Grok6 and Qwen8Max) and the Engram mechanism, a hot new technical innovation in the large model space. Its "preview" model, DeepSeek V4Lite, also generated significant buzz in the AI community earlier this year.

OpenAI announced several updates, including ChatGPT's expected 700 million weekly active users this week and new usage time limits to promote healthier engagement. The company also launched the ChatGPT Study mode for students and educators, and acquired Statsig, a leading product development data analysis platform, to enhance its product iteration capabilities.

Baidu's AI efforts include the launch of PP-OCRv6, a new OCR model with three tiers (Tiny, Small, Medium) supporting over 50 languages. Its PaddleOCR project on GitHub recently surpassed Google's Tesseract with over 3,000 stars, becoming the world's most starred open-source OCR tool. According to a report by Sullivan Consulting, Baidu Smart Cloud held a 4% market share in China's self-developed GPU cloud market in H1 2025. Additionally, Baidu's ERNIE-0-Preview-1203 model topped the LMArena Large Model Arena text ranking with 1451 points, claiming the first spot among Chinese models. At the 2026 World Robot Conference, Baidu Smart Cloud hosted a forum on embodied intelligence and signed partnerships with Beta Infinity, Zhongyi Embodied, Tuoyuan Wisdom, and Chaonao Future.

Other highlights include Mistral AI's release of MagistralSmall2, a 24B-parameter open-source inference model under the Apache 2.0 license. Step3-VL-10B, a 10B-parameter multimodal model from Step Star, showed strong performance in visual perception, logical reasoning, and math competitions. Quark updated its AI Super Box with "deep search" and "intelligent image processing" features, advancing its AI-driven content retrieval and visual creation capabilities.

Sources

• Tech 100 Net (https://www.keji100.net/)
• Machine Heart
• QuestMobile
• Sullivan Consulting
• The Information
• LMArena Large Model Arena