1X just unveiled a physics-based world model that lets its Neo humanoid robot learn new tasks by watching internet videos and reading text prompts. While the bot isn't ready for complex chores like driving, it can already pull out an air fryer basket, put bread in a toaster, and give high fives without prior demonstrations. It marks an early step toward truly self-teaching home bots.
Google just updated its Veo 3.1 video model with native 9:16 portrait outputs and up to 4K resolution. The refreshed Ingredients to Video tool lets you feed in reference photos to maintain character and background consistency across scenes without face distortion, tailored directly for short-form creators.
Kyutai just open-sourced Pocket TTS, a lightweight 100M-parameter text-to-speech model built to run entirely on CPUs without GPUs. It clocks in at roughly 6x real-time on a MacBook Air M4 with a 200ms time-to-first-audio latency, bringing voice cloning and infinite text streaming straight to consumer hardware.
YC-backed startup GRU Space just opened reservations for what it calls the first lunar hotel, asking wealthy believers for deposits between $250,000 and $1 million. The tiny two-person team plans to deploy an inflatable habitat by 2032 for SpaceX travelers. Why it matters: It sounds crazy years before infrastructure exists, but it marks an audacious early bet on commercial lunar tourism.
Anthropic just released Claude Cowork in research preview, bringing the autonomous power of Claude Code to everyday knowledge work like organizing folders, crunching expenses, and drafting reports. It runs directly inside folders you choose and keeps working even after you close your laptop, though it is currently limited to Max subscribers on the macOS desktop app.
Apple just officially teamed up with Google in a multi-year deal, choosing Gemini to power its foundational AI models and Siri's long-delayed overhaul. Reports indicate Apple could pay roughly $1B annually, while its existing ChatGPT deal remains intact. Why it matters: Apple is finally outsourcing Siri's brain to its fiercest rival.
Developer yatharthsharma3501 just open-sourced NovaSR, a microscopic 52KB audio upsampler that turns muffled 16kHz recordings into crisp 48kHz sound. Built with under 10 Conv1D layers, it clocks in at 3,600x real-time speed on a single A100 with virtually zero compute overhead. It makes on-device speech cleanup and instant TTS enhancement dirt cheap.
ShowLab just open-sourced ShowUI-Aloha, a computer-use agent that learns desktop workflows directly from human screen recordings. By turning demonstrations into reusable semantic traces instead of memorizing pixels, it hits a 60.1% success rate on the OSWorld benchmark across Windows and macOS.
PixVerse just launched R1, an interactive AI world model that streams continuous 1080P video in real time. Instead of waiting for fixed clips to render, you can inject prompts mid-stream to spawn cars or trigger background explosions on the fly. It is still glitchy and prone to hallucinations, but it shifts AI video from static outputs into live simulations.
Google just introduced the Universal Commerce Protocol (UCP), an open standard co-developed with Walmart and Shopify to let AI agents handle the entire shopping journey. The protocol introduces native Google Pay checkout inside Search and Gemini, basically turning conversational AI into an autonomous buyer. For now, agentic checkout is rolling out to eligible U.S. retailers.
Researchers just introduced RigMo, a framework that learns both skeletons and motion directly from raw 3D mesh sequences without human rigging annotations. It handles humans, animals, and even objects like fans with fewer artifacts than previous methods. Why it matters: It turns static or deforming 3D meshes into game- and film-ready animatable assets automatically.
Alibaba researchers just introduced Thinking with Map, an agentic framework that equips vision-language models with map-browsing capabilities for image geolocalization. By combining reinforcement learning with parallel test-time scaling, the model achieved a 22.1% accuracy within 500 meters—outperforming Gemini-3-Pro's map-grounded mode by 2.8x.
Researchers from Tencent just introduced VerseCrafter, a 4D video world model that turns a single image into fully controllable video. By piping 3D Gaussian paths and custom camera rigs straight from Blender into a frozen Wan2.1 diffusion model, creators can finally dictate exact scene choreography without visual degradation.
Z.ai just released GLM-Image on Hugging Face, a generative model powered by a hybrid autoregressive and diffusion decoder architecture. It handles text-to-image workflows in a single model and reached a 91.2% word accuracy score on the CVTG-2K benchmark, beating Seedream 4.5 and GPT Image 1.
Developer multimodalart just launched a Hugging Face Space that brings tactile 3D camera controls to Qwen image edits. Powered by fal's multi-angle LoRA, the demo lets you manipulate azimuth, elevation, and distance via an interactive 3D viewport or sliders. Basically, dialing in camera angles no longer requires guessing prompt keywords.
Researchers just introduced VIBE, a compact open-source image editor pairing a 2B Qwen3-VL with the 1.6B Sana 1.5 diffusion backbone. It handles 2K edits in about 4 seconds on an H100—and under 2 seconds with its CFG-distilled version—while outperforming much larger models at preserving original image details.
Sakana AI's ALE-Agent just took first place in the AtCoder Heuristic Contest 058, beating 804 human competitors in real time. Operating over four hours at a compute cost of roughly $1,300 across frontier models, the agent autonomously devised an optimization algorithm that outperformed the problem setter's intended solution. It marks the first time an AI agent has won a major competitive optimization programming contest.
UniSH just introduced a feed-forward framework that reconstructs metric-scale 3D scenes and human bodies directly from standard video. Instead of slow per-scene optimization, it recovers scene point clouds, camera motion, and SMPL bodies in one single pass, even with shaky footage. The code is now available on GitHub for local setup.
NUS researchers just built ShowUI-π, a lightweight 450M model that fixes a glaring flaw in GUI agents: smooth, continuous mouse dragging. Using flow matching, it lets agents draw, drag video timelines, and bypass rotation captchas on the fly—outperforming giants like Gemini-2.5 with a fraction of the parameters.
Tongyi Lab just open-sourced Qwen-Image-2512, an upgraded text-to-image foundation model designed to eliminate the artificial "AI look." The update renders granular human details like wrinkles and pores, sharpens animal fur, and notably boosts text rendering accuracy for complex infographics and presentation slides.
Elon Musk just announced that xAI acquired a third building named MACROHARDRR, pushing the company's training compute capacity to nearly 2GW. The massive expansion highlights xAI's aggressive infrastructure push to power the next generation of Grok models.
Runway just announced a multi-year partnership with Adobe to bring its AI models directly into Adobe's creative tools. The two companies will also co-develop specialized AI capabilities exclusively for Adobe applications. Why it matters: Creative pros will soon get Runway's generative tech right inside their daily workflows without jumping across separate tabs.
ByteDance's Seed team just introduced SeedFold, a biomolecular structure prediction model that beats AlphaFold 3 on key protein benchmarks. The team scaled the Pairformer width by 4x, introduced linear triangular attention to drop complexity to O(n²), and distilled 26.5M samples from AlphaFold 2. It marks a significant push forward in scaling laws for structural biology.
Alibaba Qwen just rolled out Qwen Code v0.6.0, introducing experimental Skills to broaden its capabilities alongside new /compress and /summary commands. The update also brings native multi-provider support for Gemini and Anthropic, plus clickable bash outputs inside VS Code. It makes Alibaba's CLI coding assistant much more versatile for everyday development.
SoftBank just sent over a final $22B tranche to OpenAI, fully wrapping up its $40B investment commitment and securing an 11% stake in the AI startup. The funds will directly bankroll OpenAI's massive compute buildout, including the Stargate data center venture with Oracle. Why it matters: SoftBank is going all-in on Sam Altman's vision as OpenAI scales toward an IPO.
Tencent just open-sourced HY-Motion 1.0, a 1B+ parameter text-to-motion model built on the Diffusion Transformer architecture and flow matching. It covers over 200 motion categories across 6 major classes and uses a complete pre-training, SFT, and RL loop to generate physically plausible 3D animations ready for typical production pipelines.
Meta is reportedly acquiring Singapore-based AI startup Manus for around $2B, marking one of the first major acquisitions of a generative AI company with substantial revenue. Manus rocketed to over $100M in ARR in just eight months with agents that handle white-collar tasks end-to-end. Meta plans to keep Manus independent while integrating its agents across Facebook, Instagram, and WhatsApp.
Fal just released FLUX.2 [dev] Turbo, an ultra-fast distilled adapter for Black Forest Labs' flagship image model. Using DMD2 distillation, it slashes inference from 50 steps down to 8, outputting 1024x1024 images in 6.6 seconds for just $0.008. While weights are free on Hugging Face, commercial use requires fal's API under non-commercial terms.
A new study from Kapwing found that over 20% of the videos YouTube recommends to new accounts are low-quality "AI slop." Researchers created a fresh profile and discovered that 104 of the first 500 suggestions were automated junk, estimating these copycat channels generate around $117 million annually. While platforms scramble to moderate, low-effort synthetic media is already dominating feed recommendations at scale.
OpenAI CEO Sam Altman just announced the company is hiring a Head of Preparedness to tackle emerging risks from advanced models. Altman noted AI is already discovering critical security vulnerabilities, making safety controls for cyber, biological capabilities, and self-improving systems an urgent priority. He warned it will be a high-stress role jumping straight into the deep end.
