Z.ai just introduced GLM-5.2, an open-source flagship model built for long-horizon coding tasks with a reliable 1M-token context. Powered by its new IndexShare architecture, it cuts per-token FLOPs by 2.9x at full context, edging past GPT-5.5 on FrontierSWE while remaining fully accessible under an MIT license.
Anthropic just analyzed 400,000 Claude Code sessions and discovered that domain know-how matters far more than programming chops. Lawyers who know contract clauses succeed at generating scripts almost as often as software engineers, with humans handling 70% of planning decisions while Claude takes on 80% of execution. Why it matters: AI coding agents are shifting leverage from pure syntax writers to people who actually understand the business problem.
Pinterest just introduced a suite of AI-powered tools ahead of Cannes Lions, including Business Assistant for Ads Manager, a standardized Model Context Protocol (MCP) integration, and an experimental visual app called Ask Pinterest. In testing, its new creative selection model boosted clicks by 7.5%. Why it matters: Pinterest is leveraging its unique Taste Graph to anchor conversational discovery in user taste and purchase intent.
Microsoft just launched Copilot Cowork into general availability worldwide, taking enterprise automation beyond chat. It independently executes long-running, multi-tool tasks to completion and runs on Anthropic’s Opus 4.8 and Sonnet 4.6 models. It requires a Microsoft 365 Copilot subscription, billed via usage-based Copilot Credits.
OpenRouter just rolled out Fusion, an API feature that queries multiple models in parallel, adds web search, and synthesizes a final response using a judge model. In testing, a budget panel of three smaller models scored nearly on par with Anthropic's flagship Fable 5 at roughly half the cost. It effectively turns model diversity into a plug-and-play research team.
Sakana AI just launched Sakana Marlin, its first commercial product designed as an autonomous research assistant for business. Users simply provide a research topic, and Marlin works autonomously for up to eight hours to map causal relationships, delivering executive summary slides and in-depth strategy reports up to 100 pages long.
Microsoft CEO Satya Nadella just published an essay urging companies to build and own their own AI learning loops. He introduced the idea of "token capital" alongside human capital, arguing firms must train agentic systems on internal data and private evals so proprietary knowledge isn't lost when switching models. Why it matters: Nadella warns that ceding all cognitive value to a handful of frontier models risks repeating the economic hollowing-out seen in early globalization.
Moonshot AI just launched Kimi Work, a desktop agent for macOS and Windows that reads your local files, drives your real browser, and runs scheduled background jobs. Powered by its K2.6 model, higher tiers can spin up a swarm of up to 300 parallel sub-agents. It is currently in testing, with subscriptions starting at $19/month alongside a free tier.
Google just rolled out its Search information agents to AI Ultra subscribers across all AI Mode languages and markets. Users can instruct AI Mode to monitor any topic, and the agent will run 24/7 in the background, pinging users with detailed summaries and web links as soon as new information breaks. It turns passive search into an active tracking assistant, with broader rollout planned for later this summer.
Coinbase just launched Coinbase for Agents, a tool that connects AI agents directly to user accounts to autonomously trade and make payments within custom limits. Available via MCP for ChatGPT and Claude or a CLI for developer terminals, it currently supports crypto spot and derivatives. Giving AI agents direct, bounded access to real capital marks a major milestone for autonomous finance.
Perplexity just integrated Deep Research into its Computer workspace, turning complex web research directly into finished PDFs, dashboards, and decks. Powered by a new "Search as Code" architecture, the model writes custom programs to run thousands of parallel search steps, pulling from internal files and premium sources like Statista and PitchBook. Why it matters: It bridges the gap between deep information retrieval and ready-to-use deliverables in a single prompt.
OpenAI is acquiring Ona to integrate its secure cloud execution tech into Codex, allowing agents to run for hours or days inside an enterprise's own infrastructure. Codex now counts over 5 million weekly users. Basically, organizations can now let agents do long-running work even when laptops are closed without compromising on security or data governance.
Sony just showed off Ace, an autonomous table tennis robot that consistently defeats ranked human professionals under official competition rules. Mounted on a heavy-duty electric rail, the bot tracks incoming ball spin within milliseconds and aggressively adjusts its own returns to force human errors. Two-time Olympic medalist Miu Hirano noted it rarely makes rally mistakes and could already compete at a top-10 women's level.
Researchers just introduced Arbor, an autonomous agent framework that turns messy trial-and-error into a structured hypothesis tree. A long-lived coordinator oversees strategy while isolated executors run tests, achieving over 2.5x the relative gains of Claude Code and Codex across long-horizon research tasks.
UC Berkeley's RDI just launched Agents' Last Exam (ALE), a demanding benchmark testing AI on real desktop workflows across 55 industries. OpenAI's GPT-5.5 topped the board with a 24.0% pass rate, narrowly beating Anthropic's new Claude Fable 5 at 22.0%. Why it matters: even the best models scored 0.0% on the hardest tasks, proving AI agents still struggle with genuine professional workloads.
HeyGen just launched HyperFrames, an official MCP connector for Claude and Claude Code that turns text prompts into animated videos. Instead of skimming dense walls of text, users can now generate code-first motion graphics, voiceovers, and subtitles directly inside Claude, exporting ready-to-share MP4s without touching a video editor.
Anthropic just added scheduled deployments and environment variable vaults to Claude Managed Agents in public beta. Agents can now automatically run recurring tasks on a cron schedule without custom scheduling infrastructure, while the new vaults inject API keys at the network boundary so the AI never sees raw credentials.
Google just teamed up with Apple to bring Gemini models into Xcode and Apple's native Foundation Models framework. Developers can now swap between on-device Apple models and cloud-hosted Gemini with minimal code changes, while Xcode gets built-in agentic coding assistance for reviewing and bug fixing. It's currently available in preview.
Google just gave NotebookLM a massive upgrade powered by Gemini 3.5, equipping every notebook with a secure cloud computer to write and execute code. It can now turn raw sources into PowerPoint slides, Excel sheets, and visual charts, while pulling web context automatically. It is rolling out to Google AI Ultra and eligible Workspace business tiers.
OpenAI just laid out a bold blueprint for its "third phase," aiming to build an automated AI researcher by March 2028 and put a personal AGI in everyone’s hands. The company stressed that human judgment must stay in control, warning against fully automated decision-making while calling for an international safety body to manage catastrophic risks.
