AI News

Follow the latest AI and automation news

28 Jun 202613 Muharram 1448 AH

Coinbase joins the rush to Chinese AI models as Western labs face a pricing stress test

Coinbase's CEO Brian Armstrong announced the company's shift to Chinese AI models like GLM 5.2 and Kimi 2.7. An automated routing system selects the best model for each request based on task and cost, improving the hit rate from 5% to 60%. Despite rising token usage, Coinbase has halved its AI spending, reflecting the pricing pressures faced by Western AI labs.

Models

28 Jun 202613 Muharram 1448 AH

Only three AI models finished above starting capital in a 500-day startup survival test

Researchers at Princeton University developed the CEO-Bench test, where AI agents manage a fictional software company for 500 days. The results revealed that most current models fail to achieve profitability, with only three AI models successfully completing the test with their initial capital. This test serves as a genuine assessment of AI capabilities in a simulated work environment, requiring models to make strategic decisions over an extended period.

Infrastructure

28 Jun 202613 Muharram 1448 AH

Optical Module MCUs Become Critical as AI Data Centers Drive Surge in 800G and 1.6T Demand

MCU chips for optical modules are now a vital semiconductor segment, with prices rising nearly 40% year-to-date. Overseas AI and optical communication firms are starting large-scale procurement of Chinese-made MCUs, leading to supply shortages. In 2025, several domestic manufacturers announced price increases, with communications sector hikes ranging from 15% to 20%, and some specialty products exceeding 50%.

💡

Peking University and DeepSeek Open-Source DSpark, Delivering Major Leap in LLM Inference Efficiency

What implications does DSpark have for future language model development?

Models

Sina's open model VibeThinker-3B aims to show reasoning compresses well but factual knowledge doesn't

28 Jun 202613 Muharram 1448 AH

Models

28 Jun 202613 Muharram 1448 AH

Sina's open model VibeThinker-3B aims to show reasoning compresses well but factual knowledge doesn't

Sina Weibo has launched the VibeThinker-3B model, featuring three billion parameters, which has outperformed models like DeepSeek V3.2 and Kimi K2.5, despite those models being up to 333 times larger in size. The effectiveness of VibeThinker-3B is attributed to its multi-stage training technique, demonstrating that logical reasoning can be compressed efficiently in smaller models, while extensive general knowledge does not compress as effectively.

Models

Inference Takes Center Stage: DeepSeek DSpark Accelerates Generation by 80%

28 Jun 202613 Muharram 1448 AH

Models

28 Jun 202613 Muharram 1448 AH

Inference Takes Center Stage: DeepSeek DSpark Accelerates Generation by 80%

DeepSeek has launched a significant update to enhance inference speed in its V4 model family, introducing the DSpark framework, which achieves up to 85% improvements in generation speed. The proactive decoding technology utilized by DSpark connects a lightweight draft model with the main model, reducing wait times for each token. This innovation enhances user experience, as inference speed directly impacts the performance of complex systems.

Models

DeepSeek DSpark Boosts Generation Speed by 85% in First Post-Funding Paper

28 Jun 202613 Muharram 1448 AH

Models

28 Jun 202613 Muharram 1448 AH

DeepSeek DSpark Boosts Generation Speed by 85% in First Post-Funding Paper

DeepSeek has launched a new framework called DSpark, enhancing generation speed by 60-85% without altering the core model architecture. Developed in collaboration with researchers from Peking University, this framework employs a speculative decoding technique to optimize performance. DSpark separates draft generation from target model verification, reducing latency. It utilizes a semi-autoregressive architecture that combines parallel networks with lightweight modules, improving token dependencies within blocks.

Business

Trump Administration Nears Approval for Anthropic's Fable 5 Model

27 Jun 202612 Muharram 1448 AH

Business

27 Jun 202612 Muharram 1448 AH

Trump Administration Nears Approval for Anthropic's Fable 5 Model

Axios reports that the Trump administration is nearing approval for Anthropic to restore access to its AI model, Fable 5. This decision comes amid increasing pressure on AI companies to clarify how they prevent misuse of advanced models, particularly concerning cybersecurity threats. Founded by former OpenAI employees, Anthropic had suspended access to its most advanced models following export restrictions imposed in June. Insiders expect these restrictions on Fable 5 to be lifted soon, with discussions continuing over the weekend.

Business

Half of Claude users say AI can already handle half their work according to Anthropic survey

27 Jun 202612 Muharram 1448 AH

Business

27 Jun 202612 Muharram 1448 AH

Half of Claude users say AI can already handle half their work according to Anthropic survey

A recent survey by Anthropic found that about half of Claude users believe AI can handle 50% or more of their tasks. The survey, which included roughly 9,700 users, revealed that 26% expect AI to cover 60% to 90% of their work within the next year. Early-career workers expressed the most concern regarding AI's impact on their jobs, while heavy users showed greater optimism about their career prospects. This disparity highlights the varying perceptions of AI's role in the workplace.

Arab AI Scene

Data and AI set to drive next phase of GCC government transformation: Artefact official

27 Jun 202612 Muharram 1448 AH

Arab AI Scene

27 Jun 202612 Muharram 1448 AH

Data and AI set to drive next phase of GCC government transformation: Artefact official

Vishal Singhal, data and consulting partner at Artefact, explains how reliable data and responsible AI contribute to tangible economic impacts in the Gulf countries. The focus has shifted from developing AI strategies to implementing large-scale programs based on robust data and effective governance. These initiatives enhance digital innovation and provide more responsive public services, supporting long-term competitiveness. Significant investments in the UAE, Saudi Arabia, and Qatar indicate that AI is no longer just a technological initiative but a national economic priority that fosters diversification and productivity.