News archive.
Earlier AI news and analysis, kept for reference. Current coverage lives in News & Analysis.
Fine-tune MoE models 3.4x faster with NVIDIA NeMo AutoModel
NVIDIA's NeMo AutoModel cuts MoE training time by 3.4–3.7x and GPU memory by 29–32% versus Transformers v5, using the same HuggingFace API. Single import line required.
Google Embeds Computer Use Into Gemini 3.5 Flash
Google moved computer use from a standalone model into Gemini 3.5 Flash, letting developers build agents that automate workflows across browsers, mobile, and desktop. Includes enterprise safeguards for prompt injection attacks.
OpenAI builds custom chip with Broadcom to control its own silicon stack
OpenAI unveiled its first custom chip as part of a deal with Broadcom, marking a shift toward owning more of its infrastructure. Here's what the move signals about frontier model economics.
OpenAI builds custom chip with Broadcom to cut AI compute costs
OpenAI designed a chip with semiconductor maker Broadcom to reduce infrastructure spending. The move signals a shift toward in-house silicon—here's what it means for model builders.
Mistral launches OCR 4 for enterprise document work
Mistral released OCR 4, expanding its AI platform beyond language models into document extraction. What the update includes and why it matters for enterprise teams handling paper-heavy workflows.
OpenAI and Broadcom build LLM inference chip together
OpenAI and Broadcom announced a custom silicon partnership focused on speeding up LLM inference. Details on performance gains and deployment timeline are not yet public.
OpenAI builds its first custom chip with Broadcom
OpenAI has commissioned Broadcom to design a custom silicon chip, signaling a shift toward in-house semiconductor control. What this means for inference costs and vendor lock-in.
GPT-5 Pro solves 3-year immunology puzzle, opening cancer research paths
Immunologist Derya Unutmaz used GPT-5 Pro to crack a mystery about T cell behavior that had stalled for three years. The finding could accelerate cancer and autoimmune disease research.
OpenAI and Broadcom build Jalapeño, a custom inference chip for LLMs
OpenAI and Broadcom unveiled Jalapeño, a custom AI chip designed specifically for large language model inference. The chip aims to improve performance and efficiency, but technical specs and benchmarks have not been disclosed.
OpenAI Paper Shows How AI Agents Handle Longer, More Complex Tasks
OpenAI research details how AI agents are expanding task complexity and productivity across roles. New capabilities signal a shift in what autonomous systems can reliably execute.
Stripe, Anthropic, OpenAI back $500M nonprofit to stop colds and flu
Intercept will fund vaccine research and air-cleaning systems to prevent respiratory infections. The effort targets the economic burden of viruses that consume 5% of human lifetime.
China's supercomputer retakes top spot from US after 9-year gap
Shenzhen's LineShine system is now the world's fastest supercomputer, dethroning California's El Capitan. But the real race isn't about raw speed anymore—it's about AI workloads.