The AllSpark team has released Iris-mini and Iris-pro, two open-source search agents built on Qwen models that lead benchmarks among open-weight models in their size classes. According to the paper, the training data and models also improved performance on tasks they were never trained for, including general tool use and office work.
A law professor spent two years testing how an AI ban, unguided AI use, and structured training affect student performance. The group without AI finished last both years. "I was wrong," the researcher writes, who had assumed that AI without guidance would do more harm than good.
In a new robotics benchmark, GPT-6 Astra shows major gains in spatial understanding. On StationeryBench, the model completed 7 out of 100 tasks with dual-arm robots, while competitor MolmoAct2 couldn't finish a single one. A researcher calls it a "step change in spatial reasoning."
Reasoning steps like calculation, formula retrieval, and deduction are clearly separable in a model's internal states, especially in the middle layers. That matters for AI safety, because models process more than their visible chain of thought reveals.
Google Research has released TimesFM-3, a forecasting model that analyzes time series alongside related data and known future events like sales promotions or weather forecasts. Instead of predicting the future step by step, the 330-million-parameter model fills in all future time points in a single pass, which cuts compute time and reduces compounding errors.
Oriol Vinyals, until recently head of research at Google DeepMind, thinks a sudden AI intelligence explosion through recursive self-improvement is unlikely. AI can speed up research by a factor of ten, he says, but it hits two bottlenecks: coming up with ideas ("research taste") and reliably judging results. Reward hacking and the speed of light add further limits. Vinyals now wants to tackle these bottlenecks with his startup Discovery Loop, co-founded with Jeff Dean, Sanjay Ghemawat, and Quoc Le.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
Traversing sparse 3D structures requires humanoid robots to perceive thin, overhanging geometry while executing agile, accurate whole-body motions. We study this problem through monkey-bar traversal, where the robot must jump to the structure, traverse it through sparse bar interactions, and land safely.
The list of obstacles that you can traverse to escape a robot is getting shorter.
9/11 was the first documented use of robots for urban search and rescue and helped create the field of disaster robotics. Personnel began assembling on the afternoon of September 11 and worked the pile from late on September 11 through October 2, when the last available robot failed. The robots found no survivors, but they located remains and helped search for routes through the rubble toward basements and stairwells where trapped firefighters might have gone.
Unitree majorly fully open-sources the UnifoLM-WLA-1.0 embodied foundation model, achieving new SOTA results across multiple benchmarks among open-source models worldwide. A single model coordinates desktop and whole-body mobile manipulation, supporting cross-task and cross-end-effector generalization, driven by one model, whole-body coordination.
Compliance is very important in physical interaction. In this work, we show how a multi-lined aerial robot uses its centroid and joint motion to achieve hybrid impedance—admittance control in contact-rich aerial manipulation tasks such as surface sliding. This work will be presented in IEEE IROS 2026.
Achieving agile and generalized legged locomotion across terrains requires tight integration of perception and control, especially under occlusions and sparse footholds. Existing methods have demonstrated agility on parkour courses but often rely on end-to-end sensorimotor models with limited generalization and interpretability. By contrast, methods targeting generalized locomotion typically exhibit limited agility and struggle with visual occlusions. We introduce a unified reinforcement learning (RL) framework for agile and generalized locomotion that incorporates a novel attention-based map encoder in the control policy.
I suspect that this demo avoids many of the things that are actually difficult about doing dishes. Not just the water and the slippery soapiness, but also identifying when a dish is dirty as well as when it is actually clean.
AI has transformed the digital world. It writes our code, generates our images, reasons in our language. But the physical world—the plants that make our power, our fuel, our steel, and chemicals—it has barely touched. ANYbotics CEO and co-founder Péter Fankhauser on the bet behind the company: Why legged robots turned out to be the way into the world’s most demanding industrial plants, what it took to certify one for explosive atmospheres after experts called it impossible, and where autonomous industrial work goes next.
Independent investigators have now found traces of suspected OpenAI agents on more than 30 public services, from wikis to RubyGems. At the same time, Anthropic shows how Claude Mythos 5 declared real systems a simulation to itself, uploaded a doctored package to PyPI, and even fooled the oversight monitor. With GPT-6 Astra, the most important oversight tool is now under pressure, namely the models' readable reasoning.
OpenAI's GPT-6 Astra tops the ErdosBench for open math problems, even though chief scientist Jakub Pachocki says math was deliberately not a priority. Instead, OpenAI is pouring resources into recursive self-improvement and alignment research. That supports the theory of an increasingly "spiky" AI development path, with extreme strength in select domains rather than broad progress, at least as long as AI can't improve itself and still needs targeted optimization with human-generated data.
Deepseek releases V4.1-Flash, a multimodal model with 552 billion parameters that cuts KV cache memory to a quarter of its predecessor. On the DeepSWE coding benchmark, it narrowly beats Opus 5 and GPT-5.6 Sol, even though only 16 billion parameters are active per token. The model ships under the MIT license and targets much cheaper AI agents.
Google Deepmind has used the AlphaGenome Atlas to predict what each of the roughly nine billion possible single-letter changes in the human genome could do. The dataset spans one petabyte, more than 30 times the size of the AlphaFold database. In one epilepsy case, the atlas helped pinpoint a previously overlooked variant as the likely cause.
The fight over an AI-generated proof of a millennium problem is heating up. Mathematician Tristan Buckmaster accuses OpenAI of academic fraud, CEO Sam Altman rejects the allegations. Terence Tao warns that cases like this could "reverse centuries of tradition in open science."
A study in Nature Biotechnology suggests that rentosertib, a drug designed with AI by Insilico Medicine, may reverse markers of biological aging. Six independent aging clocks predicted that treated patients were biologically up to six years younger than the placebo group. The trial covered just 42 patients and the drug hasn't been tested in healthy people yet.
Alibaba's research arm has released Qwen-Drive 1.0, an AI model that handles environmental perception, traffic Q&A, and route planning in one system. The researchers show that text-image models don't automatically understand three-dimensional space. Spatial awareness has to be trained on purpose. The goal is a single model that runs both the cockpit and the driving system.
Inside OpenAI, coding agents are reshaping AI research. Explore early data on agent usage, experiment velocity, task complexity, and research acceleration.
Login
