1. 【OpenAI】 has added a memory feature to ChatGPT, allowing it to automatically reference information across conversations, enhancing the personalized experience. They have also adjusted the model roadmap, prioritizing the release of o3 and o4-mini, and launched the Pioneers Program to collaborate with startups.
2. 【Google】 announced multiple AI updates at Cloud Next 2025, including the Firebase Studio development environment, Ironwood TPU chips, and a faster Gemini 2.5 Flash model.
3. 【Google】 introduced the Agent2Agent protocol to facilitate AI agent collaboration across different developer frameworks, supported by over 50 companies including Salesforce and SAP.
4. 【Samsung】 has partnered with Google to integrate the Gemini model into the Ballie home robot, planning to launch it this summer in South Korea and the United States.
5. 【Meta】 released the Llama 4 series of multimodal open-source models, with parameters reaching 10 million tokens, including MoE models with 109 billion and 400 billion parameters.
6. 【Nvidia】 launched the Llama Nemotron-Ultra 253B model, surpassing DeepSeek R1 and the Llama 4 series in inference tasks, providing open-source code and data. Nvidia also collaborated with Stanford to develop AI technology that generates coherent minute-long cartoon animations, supporting dynamic scenes and character interactions.
7. 【Amazon】 released the Nova Sonic voice AI, with a latency of only 1.09 seconds, 20% cheaper than OpenAI models, and launched the Reel 1.1 video generation AI.
8. 【Microsoft】 upgraded the Copilot application, adding memory, web browsing, and visual capabilities to compete with the Google Gemini assistant.
9. 【Midjourney】 released the V7 image generation model, improving generation quality and prompt adherence, and introduced a voice Draft mode.
10. 【Runway】 launched the Gen 4 Turbo video model, generating 10 seconds of video in 30 seconds, available for free users as well.
11. 【ElevenLabs】 introduced MCP server integration, supporting platforms like Claude to access AI voice capabilities through text prompts.
12. 【Deep Cogito】 released the Cogito v1 open-source model series, outperforming peer models through iterative distillation and amplification techniques.
