Microsoft just launched Windows AI Foundry at Build 2025, overhauling its Copilot runtime for local inference across CPU, GPU, and NPU silicon. Windows 11 is also gaining native Model Context Protocol (MCP) support so AI agents can control desktop apps directly, alongside an open-sourced WSL.
Acer Gadget just unveiled two new AI-powered wearables at Computex 2025: the titanium FreeSense Ring for health tracking with zero subscription fees, and AI TransBuds for two-way translation across 15 languages. Remarkably, only one person needs to wear the earbuds to hold a conversation.
Microsoft just started rolling out a "Hey, Copilot" hands-free wake word to Windows Insiders. As long as your PC is unlocked, you can start a voice chat without touching a keyboard or mouse. The wake word detection runs locally on-device and is opt-in, but it's currently English-only.
Stability AI just teamed up with Arm to open-source Stable Audio Open Small, a 341M parameter text-to-audio model built to run on smartphone CPUs. Powered by Arm KleidiAI, it generates up to 11 seconds of stereo sound in under 8 seconds. Why it matters: generative audio is finally moving off the cloud and directly onto edge devices for real-time sound effects.
Google just unveiled new Android security features ahead of I/O, using on-device AI in Google Messages to detect crypto, gift card, and tech support scams. Android 16 will also block sideloading during unknown calls, prevent risky accessibility permissions, and lock down stolen phones tighter against factory resets.
Apple is reportedly working on camera-equipped versions of AirPods and the Apple Watch that could launch as early as 2027. The sensors won't be for selfies or FaceTime, but are designed to feed visual real-world context directly into on-device AI models for spatial audio, gesture recognition, and visual search.
Google just detailed how it is deploying on-device AI across Search, Chrome, and Android to block evolving online scams. Desktop Chrome's Enhanced Protection mode now runs Gemini Nano locally to catch zero-day scam sites like fake remote tech support, while Android gains on-device warnings against rogue notifications, fraudulent texts, and suspicious calls.
LM Studio just rolled out version 0.3.15, bringing CUDA 12.8 support and serious performance upgrades for NVIDIA GeForce RTX users. With CUDA graphs and Flash Attention kernels baked into llama.cpp, local model throughput jumps by up to 35% across RTX 20 Series through Blackwell GPUs. The update also adds granular tool_choice controls and an overhauled system prompt editor for running offline agents.
Apple is reportedly developing custom silicon for two smart glasses models—one for daily smart tasks and another with full AR—adapted from Apple Watch tech and targeting mass production by 2026 or 2027. The company is also upgrading chips across AirPods, Watches, and enterprise AI servers to rival Meta.
Microsoft just expanded its Copilot+ PC family with a 12-inch Surface Pro and 13-inch Surface Laptop starting at $799 and $899. Both feature Qualcomm's Snapdragon X Plus chip with a 45 TOPS NPU to power local AI experiences like Recall and Click to Do on more affordable hardware.
Microsoft just unveiled a wave of new AI experiences for Windows 11 and Copilot+ PCs, headlined by an on-device agent in Settings that fixes PC issues via plain English. The update also brings dynamic photo relighting, text-to-sticker generation in Paint, and quick AI actions inside File Explorer. Microsoft is steadily weaving agentic automations directly into desktop plumbing.
Lightricks just launched LTX Video-13B, a 13-billion-parameter open-source video model available on GitHub and Hugging Face. It brings multi-scale rendering that generates high-res clips up to 30 times faster than peers, and clever quantization lets it run on consumer GPUs. Plus, it's free to license commercially for teams under $10M in revenue.
Tether CEO Paolo Ardoino just announced Tether.AI, an open-source AI runtime aimed at connecting billions of AI agents across an unstoppable peer-to-peer network. The modular platform runs on any device without API keys, integrates peer-to-peer chat app Keet, and natively supports USDT and Bitcoin payments via Tether's Wallet Development Kit. Why it matters: Tether wants autonomous agents to trade and communicate without centralized choke points.
Aurora just launched fully autonomous commercial freight runs on Texas public highways, clocking its first 1,200 driver-free miles for Uber Freight and Hirschbach. It marks the first commercial driverless trucking service operating on open roads, with plans to expand to Phoenix by late 2025.
Ai2 just released Olmo 2 1B, a 1-billion-parameter open model that outperforms Google's Gemma 3 1B and Meta's Llama 3.2 1B on math reasoning and truthfulness benchmarks. It comes with complete code and training datasets under Apache 2.0, running easily on consumer laptops and mobile devices.
Microsoft just launched Phi-4-reasoning-plus, an open-weight 14B model that punches way above its weight. Trained on curated synthetic data and GRPO reinforcement learning, it outperforms DeepSeek-R1-Distill-70B on the AIME 2025 math benchmark. Even better, it drops under a permissive MIT license and supports Ollama and vLLM right out of the box.
Vercept just launched Vy, an AI agent that runs natively on your computer, seeing your screen and directly using your signed-in apps. It is powered by VyUI, a new model that beats OpenAI, Google, and Anthropic on UI grounding benchmarks, and is available to download today.
Intel just detailed AutoRound, a post-training quantization tool designed for LLMs and vision-language models. By optimizing weight rounding and clipping ranges together, it boosts relative accuracy at INT2 by up to 2.1x over popular baselines. Quantizing a 72B model takes just 37 minutes, with direct exports to GPTQ, AWQ, and GGUF.
Alibaba just released Qwen3, a suite of hybrid AI reasoning models ranging from 0.6B to 235B parameters across MoE and dense architectures. Users can toggle thinking modes to fine-tune latency versus reasoning depth, with most models—headlined by Qwen3-32B—freely downloadable on Hugging Face and GitHub.
Microsoft just released its April preview update for Copilot+ PCs, bringing delayed AI features like Recall, Click to Do, and natural language search to general availability. Recall uses on-device NPU processing to retrace past activity with opt-in privacy controls, while Click to Do adds instant contextual shortcuts for on-screen text and images. Why it matters: After months of security overhauls, Microsoft is finally putting its headline local AI features into users' hands to prove whether PC-level ambient AI is worth buying into.
