Discover quality AI tools
Bring together the world’s leading artificial intelligence tools to help improve work efficiency
Articles published by SamuelRoberts
Chinese AI lab DeepSeek advances second round of 5 billion yuan funding targeting 50 billion valuation amid rapid growth. First seven months revenue hit 475 million yuan, ten times 2025 full year. Net loss narrowed to 715 million yuan. High gross margins of 82.9% from API calls surpass OpenAI and Anthropic, driven by popular open-source V4 models. Company raised API prices and hired investment banks for next year IPO, transitioning from low-cost open-source to scalable commercial narrative.
Chinese AI lab DeepSeek advances second round of 5 billion yuan funding targeting 50 billion valuation amid rapid growth. First seven months revenue hit 475 million yuan, ten times 2025 full year. Net loss narrowed to 715 million yuan. High gross margins of 82.9% from API calls surpass OpenAI and Anthropic, driven by popular open-source V4 models. Company raised API prices and hired investment banks for next year IPO, transitioning from low-cost open-source to scalable commercial narrative.
Tencent’s new generation large model Hy4 has had its internal test interface leaked, indicating an imminent official release. The platform follows its existing strategy of combining self-developed models with top open-source solutions, offering three core options (Hy4 for expert tasks, Hy3 as a newly upgraded general-purpose model, and DeepSeek for reasoning). Hy4 is positioned at a higher level than its predecessor Hy3, designed to handle complex business and technical challenges through either expanded parameter sizes to 1 trillion or enhanced reinforcement learning training on the existing Hy3 architecture. Hy3, released earlier this year with 295B total parameters, already outperformed similar models in tasks such as software development and financial modeling. With Hy4 set to become the core backbone for Tencent’s WorkBuddy intelligent agent application, the company aims to boost its AI capabilities in productivity areas.
Tencent’s new generation large model Hy4 has had its internal test interface leaked, indicating an imminent official release. The platform follows its existing strategy of combining self-developed models with top open-source solutions, offering three core options (Hy4 for expert tasks, Hy3 as a newly upgraded general-purpose model, and DeepSeek for reasoning). Hy4 is positioned at a higher level than its predecessor Hy3, designed to handle complex business and technical challenges through either expanded parameter sizes to 1 trillion or enhanced reinforcement learning training on the existing Hy3 architecture. Hy3, released earlier this year with 295B total parameters, already outperformed similar models in tasks such as software development and financial modeling. With Hy4 set to become the core backbone for Tencent’s WorkBuddy intelligent agent application, the company aims to boost its AI capabilities in productivity areas.
Google canceled its standalone AI Studio mobile app before launch despite 800,000 pre-orders from 168 countries. Core features will be integrated into the Gemini main app, turning it into a comprehensive hub. The web version remains for professional developers. This reflects a trend where conversation becomes development.
Google canceled its standalone AI Studio mobile app before launch despite 800,000 pre-orders from 168 countries. Core features will be integrated into the Gemini main app, turning it into a comprehensive hub. The web version remains for professional developers. This reflects a trend where conversation becomes development.
At Google I/O, the company unveiled a major upgrade to its AI creation tools, led by the new Gemini Omni model. This multimodal model integrates text, images, audio, and video inputs to generate coherent video content. A key feature is conversational editing, allowing users to describe changes like adjusting lighting or swapping characters in natural language, drastically simplifying post-production. Google aims to transform AI from a content generator into an intelligent collaboration partner for creators.
At Google I/O, the company unveiled a major upgrade to its AI creation tools, led by the new Gemini Omni model. This multimodal model integrates text, images, audio, and video inputs to generate coherent video content. A key feature is conversational editing, allowing users to describe changes like adjusting lighting or swapping characters in natural language, drastically simplifying post-production. Google aims to transform AI from a content generator into an intelligent collaboration partner for creators.
Ant Group is secretly testing an AI version of Alipay codenamed Project Bao. The new interface shifts to a conversation-first model with AI assistant Abao, enabling voice/text commands for services like food delivery and ride-hailing. The bottom bar now features Assets and Abao modules. Users can switch between original and AI versions. Limited grayscale testing requires an invitation code. The move aims to keep Alipay competitive as WeChat and Douyin advance AI agents.
Ant Group is secretly testing an AI version of Alipay codenamed Project Bao. The new interface shifts to a conversation-first model with AI assistant Abao, enabling voice/text commands for services like food delivery and ride-hailing. The bottom bar now features Assets and Abao modules. Users can switch between original and AI versions. Limited grayscale testing requires an invitation code. The move aims to keep Alipay competitive as WeChat and Douyin advance AI agents.
OpenAI launches a free ChatGPT service for verified US medical professionals including doctors and nurses. The tool aims to assist with clinical workflows and decision-making, marking a major step for AI in specialized healthcare applications.
OpenAI launches a free ChatGPT service for verified US medical professionals including doctors and nurses. The tool aims to assist with clinical workflows and decision-making, marking a major step for AI in specialized healthcare applications.
Zhipu AIs open-source lightweight model GLM-4.7-Flash surpassed 1 million downloads on Hugging Face within two weeks of release. The 8B-parameter model achieved SOTA scores for its size on benchmarks including SWE-bench and outperformed models like GPT-4o-20B, offering a cost-effective option for developers.
Zhipu AIs open-source lightweight model GLM-4.7-Flash surpassed 1 million downloads on Hugging Face within two weeks of release. The 8B-parameter model achieved SOTA scores for its size on benchmarks including SWE-bench and outperformed models like GPT-4o-20B, offering a cost-effective option for developers.
