01RoboticsInternalssingle source
Google's Gemini Robotics 2 model can control a humanoid robot's legs, torso, arms, and hands, and is the first in its robotics family to run from one set of weights, simplifying model training and design. The model achieves self-reported success rates of 45.7 to 76.3 percent on whole-body pick-ups and 32 to 92 percent on multi-finger tasks. Google also introduced a safety benchmark, ASIMOV-Agentic, and recommends running these models alongside conventional physical safety equipment. An updated Gemini Robotics reasoning model, ER 2, is paired with this release, planning steps and tracking progress from a video feed. Another model, Gemini Robotics On-Device 2, runs on the robot's own hardware without a network connection and adapts to an unfamiliar two-armed body in a few hours. The Gemini Robotics models are an experiment in learning from more generalized training data, aiming to successfully transfer knowledge from one setup to another. The release includes Gemini Robotics 2, ER 2, and On-Device 2, each with distinct features and capabilities. The models are available to early-access partners, with no public API. The performance of the models varies across tasks, with the highest success rate of 92 percent for unscrewing a light bulb and the lowest of 32 percent for sweeping with a dustpan. The introduction of the ASIMOV-Agentic safety benchmark highlights the need for caution when deploying these models in real-world scenarios. Overall, the Gemini Robotics 2 model and its accompanying releases mark a significant step forward in robotics research, with potential applications in various fields. The models' ability to generalize across different setups and tasks makes them a promising development in the field of robotics.
Provenance — who else covered this
02BusinessProductsingle source
xAI launched Grok Bot, an always-on agent with its own cloud computer, while Google introduced Gemini 3.7 Flash with significantly reduced pricing, making agent compute dramatically cheaper. Both moves validate the always-on agent category and slash inference costs, sparking a pricing war in the industry.
Provenance — who else covered this
03AgentsProductsingle source
The MCP ecosystem has grown to 22,311 servers, with new additions including RunarForge, Etch, and Sora2, and the protocol has been donated to the Linux Foundation's Agentic AI Foundation for multi-vendor governance, with a new release candidate tightening client-server contracts. The MCP SDK has also surpassed 97 million cumulative downloads.
Provenance — who else covered this
04AgentsProductsingle source
A report highlights the gap between agent demos and production reality, citing tool-call and retrieval failures as the main cause of incidents, with 88% of enterprise agents failing when deployed to real workflows. The study analyzed 73 production agent environments between January and May 2026, finding that retry loops and silent tool-call drift can lead to expensive and human-visible failures.
Provenance — who else covered this
05AgentsProduct4 sources agree
Anthropic has formalized context engineering as a discipline, positioning prompt engineering as a subset, and the community is shifting focus from saving good outputs to studying failures, with implications for agent builders and evaluation design. The concept involves curating and maintaining optimal tokens during LLM inference, and understanding failure taxonomy to inform guardrail development.
Provenance — who else covered this
06On-deviceProduct3 sources agree
Local inference infrastructure is advancing with notable performance gains, including a disk-streaming MoE approach that enables massive models to run on commodity hardware, and native vLLM runtime on RX 6000 with Windows 11. The RTX 4090 has become the benchmark consumer GPU, and practitioners highlight the importance of engineering nuance in optimizing local agent inference. Multiple models, including Qwen 30B MoE and Nemotron 3.5 Lightning 30B-A3B, demonstrate significant throughput improvements with optimized harnesses and hardware configurations.
Provenance — who else covered this
07ModelsProduct3 sources agree
MiniMax has released H3, a high-definition video generation model that accepts various input media and offers advanced editing features, but its license comes with restrictions, including territorial limitations and requirements for commercial users to display the model name, and key components of the model remain proprietary. The model has achieved top rankings in video editing and generation benchmarks, and its architecture and training data are discussed in detail.
Provenance — who else covered this
08SafetyProduct2 sources agree
A chatbot app called overit, built by researchers at Technical University of Munich and University of Cambridge, has been shown to provide relief from romantic breakup distress in a single conversation. The app uses a goal-directed approach, guiding users through four phases of conversation to challenge self-limiting beliefs and reinterpret painful memories. In a randomized, controlled trial, users who interacted with the app experienced a significant reduction in distress compared to the control group.
Provenance — who else covered this
09RoboticsProductsingle source
Hugging Face has announced a unified robotics workflow integrating Strands Agents, LeRobot library, and Hugging Face Storage Buckets, managing the entire embodied-AI lifecycle, and AWS's Strands Labs has released an open-source SDK for robot abstractions and simulation. The workflow enables developers to build agents that process visual data, interpret instructions, and perform physical actions.
Provenance — who else covered this
10AgentsProductsingle source
A new Anthropic study found that multi-agent models with conflicting goals often fail to collaborate productively, and cooperative norms don't emerge from a shared model alone. The study's findings highlight the challenges of designing multi-agent systems that can govern interactions between agents with divergent goals.
Provenance — who else covered this
11AgentsProductsingle source
New tools like Behave and MLflow's evaluation framework emerge to improve agent testing and monitoring, emphasizing the importance of evaluating the full reasoning chain, and a builder's experiment compares 8B and 24B local models as structured judges. The core issue is identifying when an agent is quietly wrong, which is a key challenge in agent development.
Provenance — who else covered this
12AgentsProductsingle source
Memory has emerged as a key bottleneck in AI agents, with new tooling addressing the issue through various approaches, including RunarForge's MCP server with session memory and entity search resolvers, and Amazon Bedrock AgentCore's layered long-term memory, all shifting towards a 'store broadly, retrieve deliberately' paradigm. These developments aim to improve memory management in AI agents by treating it as a search problem with temporal decay.
Provenance — who else covered this
13AgentsProductsingle source
OpenAI is developing a wallet for ChatGPT, enabling agents to make purchases on behalf of users, building on the Agentic Commerce Protocol co-developed with Stripe. This move raises questions about agent safety and intent verification, as the industry converges on payment rails and guardrails. Other players, such as Google, are also developing their own protocols, like Agent Payments Protocol (AP2)
Provenance — who else covered this
14RoboticsInternalssingle source
A new method using gradient descent through physics has been discovered to find gait, requiring only 500 model evaluations to achieve this from a standstill, as reported in a paper on OpenReview
Provenance — who else covered this
15ModelsProductsingle source
Z.ai released GLM 5.3, which has been praised for its capacity-to-size ratio, and Mistral is now hosting GLM 5.2 on its platform at a lower price than its own flagship model, while DeepSeek V4 Flash 0731 has been shown to be a cost-effective and high-performance model, making local and private agent orchestration viable
Provenance — who else covered this
16On-deviceInternals3 sources agree
SmallAICreator's AuroraGPT-700M-Tools brings tool-calling and web-search to a small on-device language model, while a LoRA adds tool-calling to NVIDIA Nemotron 3.5 Lightning 30B, and benchmarks show small models outperforming larger ones in certain tasks
Provenance — who else covered this
17AI securityInternals3 sources agree
Practitioners discuss the need for concrete evidence and robust security measures before deploying agents to production, with emerging tooling and techniques such as MARGINAL and VeriGuard aiming to improve agent governance and safety. The consensus pattern is defense in depth, with deterministic checks and model-based checks, as well as pre-validation on tool and model I/O.
Provenance — who else covered this
18AI securityProduct2 sources agree
A recent analysis found that code generating LLMs recommend at least 205,474 unique fake software packages, potentially leading to a significant spread of malware and security breaches across apps and companies. The rise of 'vibe' coding via tools like Cursor and Windsurf may exacerbate this issue
Provenance — who else covered this
19AI securityProduct2 sources agree
OpenAI's new Atlas browser may be susceptible to prompt injection attacks, potentially allowing malicious sites to access user data with authenticated privileges, including banking information
Provenance — who else covered this
20ModelsProductsingle source
DeepSeek-R1 achieved a 99% score on a challenging UK A level Further Mathematics exam, completing it in under 5 minutes, and its responses and the question set are available for review. The exam typically takes students 90 minutes to complete.
Provenance — who else covered this
21AI securityProductsingle source
A potential vulnerability in GPT-4 chatbots has been discovered, allowing users to access hidden prompts by asking questions without providing context, and a specific example of invoking the DALL-E tool is given
Provenance — who else covered this
22ResearchProduct5 sources agree
A recent AI study found that large language models can experience cognitive decline, including worse reasoning and poorer long-context understanding, after continual exposure to low-quality web text. The decline can lead to dumb, forgetful, and psychopathic behavior in LLMs.
Provenance — who else covered this
23ModelsProduct5 sources agree
Grok 4.6 is now available in Perplexity and Perplexity Computer, offering performance and efficiency on WANDR, and matching Fable 5 results at a lower cost
Provenance — who else covered this
24ModelsProduct3 sources agree
Alibaba Qwen has released an open-source 1 million token context model, a major recent release in the AI space
Provenance — who else covered this
25ModelsProductsingle source
The new GPT-5 pro plan has been found to perform worse than prior models, including o3 and 4o, with issues such as shorter response times and less detailed responses, and all 8 prior models have been removed
Provenance — who else covered this
26ModelsProduct7 sources agree
Gemini 3.7 Flash is now available, offering a 50% price reduction and improvements in design adherence, web development, and game creation, accessible through AI Studio, Antigravity, and the Gemini API
Provenance — who else covered this
27ResearchInternals6 sources agree
A researcher has submitted a paper on a differentiable physics engine for deep learning in robotics to the International Conference on Learning Representations
Provenance — who else covered this
28AgentsProduct4 sources agree
The idea of replacing LLM calls with agents in workflows was highlighted, and the importance of an ideal agent framework for flexibility and code ease was discussed
Provenance — who else covered this
29RoboticsProduct4 sources agree
The S2 model, a 7B-parameter open-source vision-language model, is used to power robots by processing monocular robot images and state information, including wrist pose and finger positions. This model is pretrained on internet-scale data.
Provenance — who else covered this
30AgentsProduct3 sources agree
Harnesses and evaluations help own intelligence through open agent systems, compounding loops, and governed runtimes, with harnesses and evals playing key roles
Provenance — who else covered this