AI News

Follow the latest AI and automation news

18 Jun 20263 Muharram 1448 AH

Shanghai charts IPO path for cash-hungry AI labs racing against US

Shanghai Stock Exchange has announced new rules for nonprofit AI model developers seeking to go public. Companies must have a projected market value of at least 4 billion yuan (approximately 591 million USD) to meet the listing criteria on the STAR Market. This move comes as large language model companies in China aim to attract capital amid intense competition with U.S. labs. The new standards require developers to provide evidence of their market potential, highlighting the importance of innovation in this sector.

Models

17 Jun 20262 Muharram 1448 AH

GLM-5.2 is probably the most powerful text-only open weights LLM

Chinese AI lab Z.ai released the GLM-5.2 model on June 13, followed by open weights on June 16. This model boasts 753 billion parameters and a context window of one million tokens, enhancing its capabilities over previous versions. The model has generated significant buzz, being ranked as the leading open weights model on the Artificial Analysis Intelligence Index, surpassing competitors like MiniMax-M3 and DeepSeek V4 Pro, showcasing its performance advancements.

Agents

17 Jun 20262 Muharram 1448 AH

OpenAI Launches Scheduled Tasks Feature for ChatGPT

OpenAI has launched a new update for ChatGPT that includes a scheduled tasks feature, allowing users to set reminders and automate repetitive tasks. The rollout of this feature begins today, enhancing task management capabilities. This new feature aims to improve notification mechanisms and simplify task access, thereby enhancing user experience and increasing the chatbot's effectiveness in handling daily tasks. This addition is expected to facilitate better organization of work.

💡

Amazon, Nvidia, and AMD bet $310 million on AI startup building 3D world models

What impact will world models have on the future of AI?

Business

WakeCap acquires Frontline as it builds an end-to-end construction intelligence platform

17 Jun 20262 Muharram 1448 AH

Business

17 Jun 20262 Muharram 1448 AH

WakeCap acquires Frontline as it builds an end-to-end construction intelligence platform

WakeCap, a construction technology company, has acquired Frontline, an AI-powered construction planning software provider. This acquisition aims to expand WakeCap's capabilities from real-time jobsite monitoring to project planning and optimization. Founded by Luis Martinez and Ricky Ding, Frontline offers AI-driven tools that help project teams optimize schedules through simulations and stress testing. The acquisition will enable WakeCap to create an integrated intelligence platform linking project plans with real-time field execution data.

Models

World model maker Odyssey nabs $1.45B valuation backed by Amazon and other big names

17 Jun 20262 Muharram 1448 AH

Models

17 Jun 20262 Muharram 1448 AH

World model maker Odyssey nabs $1.45B valuation backed by Amazon and other big names

Odyssey has announced its position as a leading startup in the realm of AI models, moving beyond large language models. World models represent a new frontier in AI development, enabling a deeper understanding of data and contexts. These models open up new possibilities for AI applications, enhancing companies' innovation capabilities. Odyssey's advancements push the boundaries of technology, indicating that businesses adopting these models will achieve significant progress across various sectors, unlocking new market opportunities.

Models

Zhipu AI's GLM-5.2 closes in on closed-source leaders in coding marathons

17 Jun 20262 Muharram 1448 AH

Models

17 Jun 20262 Muharram 1448 AH

Zhipu AI's GLM-5.2 closes in on closed-source leaders in coding marathons

Zhipu AI has unveiled its GLM-5.2 model, featuring a stable context of one million tokens under the MIT license. In the FrontierSWE benchmark for extended coding tasks, GLM-5.2 trails Anthropic's Claude Opus 4.8 by just one percentage point, showcasing its competitive edge. However, the model still lags behind closed-source competitors in reasoning capabilities, highlighting ongoing challenges in enhancing its performance in this critical area.

Agents

Agent-to-Agent (A2A) Protocol: How Cross-Agent Communication Works in Production

17 Jun 20262 Muharram 1448 AH

Agents

17 Jun 20262 Muharram 1448 AH

Agent-to-Agent (A2A) Protocol: How Cross-Agent Communication Works in Production

Google launched the A2A protocol in April 2025 as an open standard for communication between intelligent systems. The protocol utilizes HTTP for data transmission and JSON-RPC for structured messaging, facilitating collaboration among smart agents. This design supports asynchronous operations, allowing one agent to send tasks to another while continuing its own work. The A2A protocol aids developers in avoiding custom coding for tool integration, thereby reducing reliance on a single vendor. It aligns with security and authorization patterns used in organizations, simplifying framework replacements without impacting infrastructure.

Agents

Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it.

17 Jun 20262 Muharram 1448 AH

Agents

17 Jun 20262 Muharram 1448 AH

Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it.

Physical AI faces significant data challenges that hinder its ability to achieve milestones similar to those of large language models. Effective mechanisms for data collection and analysis are essential for these systems to learn and adapt. Without addressing these challenges, the advancement of physical AI may lag behind that of large language models.

Agents

Nvidia research shows robots that train themselves through AI coding agents

17 Jun 20262 Muharram 1448 AH

Agents

17 Jun 20262 Muharram 1448 AH

Nvidia research shows robots that train themselves through AI coding agents

Researchers from Nvidia, Carnegie Mellon University, and UC Berkeley are leveraging AI coding agents to teach robots dexterous grasping skills in real-world scenarios. A fleet of eight robots has achieved up to 99% success on complex tasks requiring precision. This study marks a significant advancement in developing self-learning robots, utilizing AI techniques to enhance performance in intricate tasks. The robots rely on AI coding agents for self-training, showcasing notable progress in robotics.