ARC-AGI-3 tests real-time intelligence in games, Nemotron leads in distilled reasoning, and Manus shows how context engineering boosts agent performance without new models.
New research challenges AI reasoning vs memorization • UK launches 21-exaflop Isambard-AI supercomputer • Adobe Firefly adds video/audio tools • NVIDIA's 2.5B ASR model.
LTX-Video enables real-time AI video creation, Asimov redefines deep code understanding, and Google adds Gemini 2.5 Pro & Deep Search to make Search smarter than ever.
OpenAI’s ChatGPT Agent acts as a smart assistant, Stanford launches fully open Marin model, and Anthropic’s Claude reshapes financial analysis with real-time data tools.
Mistral's Voxtral redefines voice AI, Google boosts AI-powered cybersecurity, and Amazon S3 Vectors launches native vector storage for scalable AI and semantic search.
Grok adds flirty anime AI avatars on iOS, Claude now designs in Canva via chat, and Google’s NotebookLM debuts featured notebooks with expert-curated, shareable content.
Kimi K2 launches as an open-source agentic AI, Kiro redefines coding with spec-driven dev, and Claude now connects to tools like Notion, Figma & Stripe for smarter workflows.
Microsoft launches Phi-4-mini-flash for mobile AI, Nvidia unveils export-safe Blackwell chip for China, and Meta acquires PlayAI to boost voice tech in AI characters and tools.
Mistral's Devstral hits 61.6% on SWE-Bench • Claude integrates with Canvas & education platforms • Reka open-sources Flash 3.1 with 21B params. AI coding & learning evolve fast!
Google’s Circle to Search adds AI Mode for deeper insights. New Flourishing AI Benchmark evaluates ethical AI. McDonald’s AI hiring bot suffers major data breach.
Musk launches Grok 4 amid controversy, AI2 unveils FlexOlmo for private collaborative training, and Google debuts T5Gemma—an efficient, flexible encoder-decoder model.
Comet redefines browsing with AI, SmolLM3 brings powerful multilingual reasoning in a tiny model, and OpenAI prepares to launch an AI-first browser to rival Chrome.
