Phi 4's reasoning process shown in screenshot by Simon Willison.

OpenAI has been developing reinforcement fine-tuning (RFT) since last December, and it is now available for verified organizations using the o4-mini model. RFT uses a programmable evaluation system, ...

A new report highlights how Chinese AI startup DeepSeek is outperforming Western benchmarks with a research team that is almost entirely China-based—and what that means for the US’s position in global ...

OpenAI is reportedly in discussions with the U.S. Food and Drug Administration (FDA) about using AI to help evaluate pharmaceuticals, according to a new report from Wired. At the center of the talks ...

Alibaba’s research lab Tongyi has introduced ZeroSearch, a new method for training large language models to handle search tasks—without relying on real web searches. For chatbots to answer questions ...

Google is now using AI models to protect Chrome users from online scams. On desktop, the company has rolled out its local Gemini Nano language model to quickly spot fraudulent websites, including ones ...

Hugging Face has released a new open-source AI agent designed to handle basic computer tasks. For now, though, it feels more like a shaky tech demo than a useful assistant. The Open Computer Agent, ...

IBM CEO Arvind Krishna says the company has used AI and so-called AI agents to cut several hundred jobs in its HR department. At the same time, IBM has created new positions in areas like software ...

Anthropic has launched a web search feature for its Claude API, letting developers combine Claude models with up-to-date web data without building their own search infrastructure. Claude decides when ...

ByteDance introduces Seedream 3.0, a new text-to-image model. Benchmarks suggest improvements over GPT-4o and Midjourney in speed, accuracy, and visual quality. ByteDance has released Seedream 3.0, a ...

OpenAI has made its multimodal image generation model, GPT-Image-1, available to developers through the API. Previously limited to ChatGPT, the model is now being adopted by companies including Adobe ...

Microsoft is adding support for the open Agent2Agent (A2A) protocol to Azure AI Foundry and Copilot Studio, aiming to enable AI agents to work together across different platforms. With A2A support, ...