X just rolled out a completely rebuilt Android app after a year-long engineering effort. The new version fixes long-standing lag, improves scrolling smoothness, and brings faster notification delivery. Hosting Spaces and optimizations for older devices are still in the works, but the cleaner foundation promises faster updates ahead.
Nvidia just unveiled Cosmos 3 Edge at SIGGRAPH, packing a 4-billion-parameter world model designed to run physical AI locally in real time. The company also launched a Synthetic Video Detector NIM microservice to help newsrooms spot AI fakes frame by frame, alongside MCP integrations linking AI agents directly into Adobe, Blender, and Unreal Engine. It pushes both physical reasoning and workflow agents straight onto local hardware.
xAI just open-sourced Grok Build, releasing the harness behind its coding agent and terminal UI on GitHub. The release details how it handles context assembly and tool dispatch, while offering plug-in support for MCP servers and subagents. Best of all, developers can compile it themselves and run it fully local-first with their own inference.
PrismML just released Bonsai 27B, an open-source multimodal model based on Qwen3.6 that fits onto consumer phones and laptops. Its extreme 1-bit binary variant shrinks the memory footprint to just 3.9GB—roughly 14x smaller than standard FP16—while hitting 11 tokens per second on an iPhone 17 Pro. Both the 1-bit and 1.58-bit ternary models are freely downloadable under the Apache 2.0 license.
Apple just brought Siri AI to the watchOS 27 developer beta, delivering a unified assistant experience across the watch, iPhone, and Mac. The upgraded assistant seamlessly surfaces reminders and notes on the fly while syncing query history across devices. It finally bridges the smartwatch gap, though slight latency and AI quirks remain.
Clearcam just made upgrading home security painless by turning regular RTSP camera feeds into AI-powered smart systems. Powered by Qwen3-VL, it brings real-time object tracking, incident summaries, and natural language footage search to Mac and Linux. Basically, you get advanced surveillance intelligence without replacing a single piece of hardware.
Open-source matting model Lucida just launched to fix where typical removers fail, cleanly isolating semi-transparent glass, camouflage, drop shadows, and fine hair. At under 900MB on an MIT license, it runs locally on consumer GPUs and beats commercial baselines across several edge cases.
ChatCut just introduced its desktop app with local MCP support, enabling agents like Codex and Claude Code to handle AI video editing. It allows agents to recreate footage while swapping people or scenes and preserving motion timing. The tool is available on macOS and Windows, with Linux still unsupported.
Nvidia just updated its open-source PID decoder to version 1.5, bringing sharper detail and cleaner color fidelity without grid artifacts. It turns 512×512 latents straight into 2K or 4K renders on consumer GPUs in mere seconds. Why it matters: it bypasses slow, two-step upscaling workflows and works out of the box with models like FLUX and Qwen-Image.
Qualcomm researchers just open-sourced MobileWan, compressing Alibaba's 5B-parameter video model to run entirely on commercial smartphones. By pruning attention heads and slashing diffusion to three steps, it generates a 5-second 480p clip in about 20 seconds without touching the cloud.
Apple just confirmed that its new Apple Home AI capabilities will require at least a $10-per-month 2TB iCloud+ plan or the top-tier Apple One Premier. The feature locally analyzes HomeKit Secure Video footage for people, objects, and events to build searchable highlight summaries. Regular notification grouping and 4K support remain available on cheaper tiers.
Firefly Aerospace just revealed plans to deploy NVIDIA’s Jetson edge AI platform into lunar orbit during its Blue Ghost 2 mission in late 2026. Instead of beaming massive raw sensor feeds back to Earth, onboard Jetson modules will process imagery directly in space, slashing data turnaround from weeks to near real time.
Baidu just introduced Unlimited OCR, a 3B MoE document parser that stops memory bloat across 40+ pages. By using an R-SWA mechanism to cap the text KV cache to the last 128 tokens, the model processes long documents at a completely constant speed.
InternScience just open-sourced Agents-A1, a 35B mixture-of-experts model built specifically for multi-step agentic workflows and tool use. Despite its compact footprint, it rivals frontier giants like GPT-5.5 on science and agent benchmarks, with quantized versions small enough to run locally on consumer hardware.
Liquid AI just rolled out two 350M-parameter retrieval models, LFM2.5-ColBERT and LFM2.5-Embedding, bringing fast multilingual search across 11 languages to edge devices and CPUs. The twin models tackle both single-vector speed and token-level accuracy, outperforming rivals on NanoBEIR and MKQA benchmarks. Why it matters: developers can now run high-accuracy cross-lingual search locally via GGUF without blowing up compute budgets.
Qualcomm just unveiled the Snapdragon Reality Elite, packing a 160% jump in NPU performance and 60% faster graphics for next-gen headsets. The chip debuts inside Google and Xreal's upcoming Aura glasses this fall, aiming to run Gemini Live multimodality right on device. Preorders are open now.
Moonshot AI just launched Kimi Work, a desktop agent for macOS and Windows that reads your local files, drives your real browser, and runs scheduled background jobs. Powered by its K2.6 model, higher tiers can spin up a swarm of up to 300 parallel sub-agents. It is currently in testing, with subscriptions starting at $19/month alongside a free tier.
Google just open-sourced DiffusionGemma, an experimental 26B MoE model that ditches token-by-token generation for text diffusion, stamping out 256 tokens in parallel. It delivers up to 4x faster GPU inference, clocking over 1,000 tokens per second on an H100 and fitting into 18GB VRAM when quantized. Why it matters: It drastically cuts latency for local inline editing and code infilling, though Google notes raw output quality still trails standard Gemma 4.
Princeton researchers just released i1, a 3B diffusion model that openly shares its weights, code, and entire data curation pipeline. While not built to outperform massive closed models, it runs easily on consumer hardware and gives anyone a complete, transparent recipe to train image models from scratch.
Anthropic just launched a Swift package integrating Claude with Apple's Foundation Models framework across iOS 27 and macOS 27. Developers can now run quick tasks on-device and cleanly hand off multi-step reasoning, coding, and web searches to Claude via typed Swift outputs. Why it matters: it seamlessly unites local privacy and cloud horsepower inside native apps.
