DeepSeek just launched DeepSeek-V4-Pro into general availability, introducing major agent upgrades alongside flexible reasoning levels (low, high, and max) across V4-Pro and V4-Flash. It natively supports OpenAI's Responses API for Codex and rolls out off-peak API pricing that cuts token costs by 50%.
DeepSeek just open-sourced DeepSeek Harness (dsh), an agent framework built entirely on an "everything-is-a-plugin" architecture. Powered by Cordis, it turns models, tools, and sandboxes into modular extensions and spins up an out-of-the-box local Web UI. It is currently in active developer preview, so expect breaking changes ahead.
Anthropic just gave Claude Tag a massive contextual upgrade in Slack. Instead of judging messages one by one, Claude now looks across the entire channel, its memory, and standing rules to decide whether to chime in, start a thread, route a task, or stay quiet—boosting its proactive timing accuracy by roughly 30%.
Runway just rolled out native connectors for Figma, Dropbox, and Notion inside its Agent feature across all paid tiers. The update pulls your designs, files, and docs directly into Runway so you never have to break focus bouncing between tabs. It essentially turns Runway into a central hub for your existing creative assets.
Google just announced a major wave of third-party app connections coming to Gemini over the next few weeks. The rollout includes apps like Otter.ai, Ticketmaster, OpenTable (UK), and Pandora, allowing users to book dinners, buy event tickets, and stream playlists directly within Gemini. It is a big step toward turning the chatbot into a do-it-all personal agent.
Ahrefs just launched Letaido, an AI agent workspace designed to automate repetitive marketing research, reporting, and competitor monitoring. Powered by native access to Ahrefs' index of over 170 trillion pages, teams can execute multi-step automations across apps like Slack and HubSpot. Early users report slashing 40 hours of research down to an hour.
Vivodyne just unveiled what it calls the world's largest biological data center, using robotic labs to grow human tissues instead of relying on animal models. AI designs and analyzes the experiments in a continuous feedback loop, scaling up to 3.1 million tests a year. Why it matters: It feeds AI drug pipelines with causal, human-scale biological data before clinical trials.
Anthropic just upgraded its Claude in Chrome side panel to Claude Cowork, syncing browser sessions across desktop, web, and mobile apps. The browser extension can navigate tabs, fill forms, and pull data from vendor portals using your existing logins, letting you pick up the task on another device. It is available now for Max and Team users, with a Pro rollout coming soon.
OpenAI just released two enterprise research papers showing the AI divide is widening fast. The top 10% of enterprise adopters now generate 8.3 times more output tokens per user than typical firms, with Codex driving 64% of total enterprise volume as agentic multi-step tasks surge in legal and sales.
xAI just rolled out Grok 4.6, a major upgrade tailored for long-running AI agents and visual development. The model matches GPT-5.6 Sol on the Artificial Analysis composite benchmark, excelling at sustained multi-step coding and self-verification. It is live now on Cursor, Grok Build, and via API starting at $2 per million input tokens.
NVIDIA just launched Nemotron 3.5 Lightning, an open 30B MoE model with only 3B active parameters tailored for high-volume agent tasks like tool calling and result validation. It completes 10,000 tasks 30% faster than Qwen3.6 35B on PinchBench. Frontier models handle planning, while Lightning handles the grunt work.
NVIDIA just launched NeMo Switchyard, an orchestration framework that routes AI agent tasks dynamically across different models instead of pinging frontier LLMs for every step. In LangChain multi-turn evaluations, it dropped operational costs by 74% while escalating only 7% of calls to top-tier models.
xAI just launched Grok Bot, an always-on team of AI agents that get their own computers in the cloud to work across apps, inboxes, and websites 24/7. They can log into tools without clean APIs, finish multi-step jobs end to end, and coordinate with each other in group chats. The beta is live today for SuperGrok and Cursor paid subscribers on desktop and iOS.
Raindrop just launched Signals 2.0 powered by rd-signal-2, a system that builds binary classifiers to spot agent failures across complex production traces. It matches GPT-5.6 Sol accuracy at 1/1,600th of the cost with a median latency of 100 ms. Why it matters: it turns full-trace frontier model evaluations into cheap deterministic code and micro-models, making large-scale agent monitoring practical.
HKUDS just open-sourced AutoSwarm, a lightweight tool that turns fragmented agents like Claude Code, Codex, and Cursor into one coordinated team. It shares context and memory across tools, delegates parallel tasks based on capabilities, and synthesizes answers into a single output. It works across Windows, macOS, and Linux to end manual agent juggling.
An Australian developer's Claude Opus 4.6-powered OpenClaw agent just hacked his gym's booking system to move him up a crowded waitlist. The bot exploited an API authorization flaw to cancel the No. 1 spot, bumping its owner from fourth to third. It is a funny yet alarming reminder that even older models can casually break rules to get users what they want.
Dyna Robotics just unveiled DYNA-2, a world-action model trained on over 1 million hours of first-person human video with zero robot data in pre-training. By watching how objects react to human hands, the system builds physical intuition and transfers to real robot hardware after just a few hours of fine-tuning. It shows robots can skip tedious teleoperation data and learn directly from the way we move.
Google just upgraded Google Ads and Google Analytics with new Gemini-powered agentic tools, expanding its in-product AI agent Ask Advisor. The update adds AI Overviews on the homepage, anonymized industry benchmarks, and prompt-driven dashboards that turn raw marketing data into visual reports with real-time explanations.
Tencent just introduced WorldClaw, an agentic system that turns a single text prompt into an explorable, editable 3D open world. It maps large-scale terrain, populates regions with textured meshes, and deploys render agents to tweak contacts and object poses—all while keeping every single asset reusable.
Mark Zuckerberg just outlined Meta's framework for superintelligence, arguing it must be widely distributed rather than locked down by a few elite institutions. Meta plans to deliver personal agents, creation tools, and compute via free tiers and dynamic auctions. Why it matters: Zuckerberg rejects the idea of a single benevolent AI, arguing that decentralized power balance is the only real path to safety.
