Claude Artifacts Hit Pro & Max, Fable 5 Tops Remote Labor Index, ZCode Launches

Anthropic brings Claude Code's live Artifacts to every Pro and Max subscriber, Fable 5 quietly doubles its own automation lead, Z.ai ships a free agentic coding environment for GLM-5.2, and Anthropic starts talking chips with Samsung. Blend Roundup 2026-07-03 https://x.com/ClaudeDevs/status/2072770790114914317 https://x.com/CAIS/status/2072360965522489789 https://x.com/Zai_org/status/2072349453361557898 https://x.com/TechCrunch/status/2072750315590533518 Anthropic brings Claude Code's live Artifacts to every Pro and Max subscriber, Fable 5 quietly doubles its own automation lead, Z.ai ships a free agentic coding environment for GLM-5.2, and Anthropic starts talking chips with Samsung - let's get into today's AI news. Claude Code can now build and publish live web pages for anyone on a Pro or Max plan, not just Team and Enterprise accounts. Ask it for an artifact - a chart, a small app, a working prototype - and Claude writes the code, then publishes it straight to a private page on claude.ai. That page keeps updating in real time as Claude keeps working, so you're watching the finished thing take shape rather than waiting for a final export. Previously, artifacts were something you could preview inside a chat and not much else. Now they're a genuine deployment target baked into the coding workflow itself - no separate hosting step, no manual redeploy. It closes the gap between Claude writing code and that code actually running somewhere you can use it. New Remote Labor Index results put Claude Fable 5 at the top of the field by a wide margin - it now completes 16.1% of real remote-work projects at a professional standard, tested across 240 projects in 23 domains. That's roughly double the next model, and nearly four times Opus 4.6's 4.2% automation rate from earlier this year. The benchmark isn't asking whether a model can pass a coding test - it's asking whether its actual deliverable would satisfy a client who hired a professional. Sixteen percent still means most remote work stays firmly human for now, but the frontier has more than quadrupled in under eight months. Z.ai has launched ZCode, the official development environment for its GLM-5.2 model, going up directly against Cursor, Claude Code and GitHub Copilot. It's less an IDE than an agent-first workspace - a file manager, terminal, git panel and live browser preview all built around the agent conversation, free to download on MacOS, Windows and Linux. Existing GLM Coding Plan subscribers get one and a half times their usual usage quota inside it, and it supports bringing your own key if you'd rather use your existing subscriptions. The pitch is familiar - one agent, one workspace, minimal context-switching - but Z.ai is betting GLM-5.2's long context window makes it stick. Anthropic is in early talks with Samsung about building a custom AI chip, centred on Samsung's 2 nanometre process and its advanced packaging lines. Nothing's decided yet - not whether it's for training or inference, not performance targets, not even a firm commitment to build it. But Anthropic has been quietly hiring hardware engineers, including one poached from OpenAI's own chip team last month, and this follows a now-familiar pattern - OpenAI, Google and Amazon all run custom silicon alongside their Nvidia spend. Anthropic insists AWS Trainium, Google's TPUs and Nvidia GPUs stay central to its compute strategy regardless. Diversifying your suppliers is just good procurement, whatever the press release says.