Sakana AI just launched Fugu-Cyber, an orchestration model scoring 86.9% on the CyberGym benchmark to rival top frontier models. Instead of relying on a single LLM, it routes enterprise defense tasks across a dynamic pool of specialized security agents. Access is currently available via application review on their API endpoint.
Browser-use just merged a new feature into browser-harness that records browser agent sessions and automatically edits them into Screen Studio-style clips. The agent captures per-action screenshots, hides sensitive credentials, and renders smooth zooms, synthetic cursors, and captions inside the browser without needing ffmpeg.
Runway just launched Agent 2.0, turning simple text prompts into complete marketing briefs and multi-platform campaign assets. It can ingest performance data from platforms like Meta and YouTube, then orchestrate models like Gen-4.5 to output auto-formatted variants for different markets. Why it matters: Runway is making a serious play for end-to-end marketing workflows, moving far beyond standalone video generation.
Nvidia and Stanford researchers just unveiled RoboTTT, scaling a robot's visuomotor context to 8,000 timesteps—roughly five minutes of continuous real-time control. By using Test-Time Training to compress action history into fast weights without adding inference latency, the robot aced a 10-stage assembly task where traditional baselines failed. Why it matters: Context length is emerging as the next big scaling law for physical AI.
NVIDIA just released Nemotron 3 Embed, a trio of open-weight models aimed at production RAG and agentic workflows. The 8B flagship took first place on the RTEB multilingual benchmark, while a 1B variant gets native NVFP4 acceleration for Blackwell chips. Beyond 32k context and code retrieval, the key hook is efficiency: cleaner context retrieval directly cuts down multi-turn agent reasoning costs.
Google just rolled out connected apps inside Search’s AI Mode for U.S. users, linking external services directly into search results. You can now add recipe items straight to an Instacart cart, pull up Canva templates, or save curated playlists to YouTube Music without switching tabs. Google plans to expand to more partner apps soon.
1Password just rolled out an integration that lets Claude automatically fill credentials to finish multi-step web tasks without ever seeing your actual passwords. Using a new zero-exposure framework, credentials and MFA codes are injected through a secure channel hidden from Anthropic's models, requiring only a quick biometric sign-off per task. It solves a major security roadblock for autonomous browser agents.
Roblox just unveiled Build, a mobile-first AI creation tab inside its app that lets anyone generate playable 3D games directly from plain text prompts without writing code. The tool automates mechanics, environments, characters, and audio using proprietary 3D gaming models. It launches a free public alpha in New Zealand on July 28, with professional testing and analytics agents coming to Studio soon.
Moonshot AI just unveiled Kimi K3, a 2.8-trillion parameter model claimed as the world's first open 3T-class frontier model. Built on a sparse MoE activating 16 of 896 experts, K3 features native vision, a 1-million-token context window, and high-level autonomous coding skills. Full model weights are set to drop on July 27, 2026.
xAI just launched Automations in Grok, letting the AI run repetitive tasks on a schedule or automatically whenever matching emails arrive. Each execution opens a real conversation thread with saved history, tool connections, and custom reporting options. Scheduled tasks are free for everyone, while email triggers require a SuperGrok subscription.
xAI just released Grok 4.5, its flagship model tailored for coding, agentic workflows, and knowledge work. Trained alongside Cursor on tens of thousands of NVIDIA GB300 GPUs, it delivers 80 TPS generation speed and double the token efficiency of rivals, priced at $2 input and $6 output per million tokens. It is currently available with temporary free access in Grok Build and Cursor.
DoorDash just launched dd-cli in limited beta, a command-line tool that lets AI agents search stores, hunt down deals, and complete checkout autonomously. It marks a push into agentic commerce where software handles physical transactions directly. Early access is currently open to macOS developers in the US and Canada via a waitlist.
OpenAI just dropped Codex Micro, a $230 physical macropad built with Work Louder to manage its Codex coding platform. The hardware packs 13 mechanical keys, a dial, and a joystick to monitor agent thread statuses via colored lights, tweak reasoning levels on the fly, and approve code changes. All controls are remappable through the ChatGPT desktop app.
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.
Perplexity just launched SPACE, a purpose-built sandbox platform powering its Computer agent system. SPACE runs tasks inside ephemeral Firecracker microVMs that self-destruct once finished, keeping sensitive credentials strictly outside until the precise second of use. Rolling snapshots capture live memory every minute, letting users pause and resume complex agent workflows up to a week later.
Thinking Machines just launched Inkling, an open-weights MoE model featuring 975B total and 41B active parameters alongside a 1M token context window. Pretrained across text, images, and audio, it lets developers dial thinking effort up or down to balance cost against latency. Why it matters: Open-source builders finally get a frontier-scale multimodal base designed from the ground up for agentic fine-tuning.
BaoCut just turned tedious video editing into single-line agent commands. Hooking directly into Claude Code or Codex, it automatically transcribes footage, strips filler words, and translates subtitles directly inside an editable timeline. Why it matters: Creators no longer need to scrub audio wave by wave just to fix a single bad take or subtitle line.
Perplexity just rolled out Brain for Computer, a self-improving memory layer that connects context across past sessions to boost accuracy by 25% while trimming costs by 13%. The update also introduces Claude Fable 5 as a multi-step task orchestrator, allows mid-task model switching, and adds one-click web publishing to pplx.app or custom domains.
Anthropic just gave the Claude Code desktop app its own in-app sandboxed browser. Claude can now pull up live docs, design mockups, or any site to read, click links, and interact just like it does with local dev servers. Sessions can also be configured to persist between prompts.
OpenAI just published a paper proving the cycle double cover conjecture, a graph theory problem unsolved for over 50 years. Powered by GPT-5.6 Sol Ultra, the model spun up 64 subagents and generated the entire proof in under an hour. Why it matters: Frontier AI is officially shifting from helping mathematicians check steps to independently cracking open problems.
