AI Updates This Week: 12 Shocking Wins Over ChatGPT & Claude

If you only read one roundup of AI updates this week, make it this one. Because buried inside these 12 announcements is a genuine shift in power: a free tool just combined ChatGPT and Claude into something smarter than either model alone, Meta reportedly told its own engineers to stop using Claude Code, and Anthropic’s banned “super model” quietly came back — but not in the shape anyone expected.

Here’s the full breakdown of the AI updates this week that actually matter, why they matter, and what’s genuinely new versus what’s just noise.

Why AI Updates This Week Matter More Than Ever

For two years, the entire AI conversation has revolved around one argument: ChatGPT or Claude, which is smarter, which is worth paying for. This week broke that framing entirely. A free tool proved you don’t need to pick a side — you can combine both. Meanwhile, a wave of free, open-source Chinese models is closing the gap on premium U.S. tools faster than most people realize, and it’s forcing every major lab to rethink pricing, access, and even who gets to use their own products internally.

The Top 12 AI Updates This Week You Need to Know

1. Hermes’ “Mixture of Agents” Merges ChatGPT and Claude Into One AI

A multi-model platform called Hermes rolled out a feature called Mixture of Agents that lets you pick any two AI models — say, ChatGPT and Claude — run them on the same prompt simultaneously, and have a third pass merge the strongest parts of each into a single, cleaner answer. On Hermes’ own benchmark, the combined output scored roughly 8% above Claude’s Opus and 11% above GPT-5.5 individually. The takeaway from this AI update this week is simple: you no longer need exclusive access to one flagship model to get flagship-level output.

2. Meta Reportedly Bans Its Engineers From Using Claude Code and Codex

According to a report from The Information based on internal documents (unconfirmed by Meta), Meta has told its engineers to stop using Claude Code and OpenAI’s Codex. The reported reason isn’t security — it’s distillation risk. If Meta’s internal AI trains on code generated by a rival model, Meta could be accused of copying a competitor’s intellectual property without permission, a legal exposure that Anthropic has already raised against Chinese firms like Alibaba and DeepSeek. Meta is reportedly steering its teams toward its own internal coding tools instead.

3. A Whisper-Powered Smart Ring Lets You Type Without a Keyboard

A new wearable ring lets you dictate text to your computer just by whispering, using the dictation engine WhisperFlow, known for picking up extremely quiet speech. The ring also doubles as a tiny trackpad, letting you swipe to edit, delete, or move your cursor without ever touching a keyboard. It’s available for pre-order now, with shipping expected near the end of the year.

4. Z.ai Launches Zcode, a Full Coding App Built Around GLM 5.2

Chinese AI company Z.ai released Zcode, a dedicated coding environment built around its free, open-source GLM-5.2 model — one that already trades blows with Claude’s Opus tier. Previously, using GLM meant plugging it into a third-party tool; now it has its own native app for Mac, Windows, and Linux. Zcode’s biggest strength isn’t raw coding power (GLM 5.2 ranked second in a recent coding benchmark, behind GPT-5.5) — it’s flexible model-switching, letting you plug in almost any model, including Claude Sonnet 5, through a single API key.

5. Anthropic Launches Claude Sonnet 5

Anthropic quietly shipped Claude Sonnet 5, its new mid-tier model and, by Anthropic’s own framing, its most agentic yet — meaning it can plan a multi-step task and work through it independently rather than just answering a single prompt. It performs close to Anthropic’s flagship Opus tier at a lower cost, and if you’re already a Claude user, you may already be on it by default.

6. Claude Fable 5 Returns After Its Export-Control Ban — But Weaker

This is the biggest of this week’s AI updates this week: Claude Fable 5, Anthropic’s most powerful model, launched and was pulled offline outside the U.S. within days after the U.S. Department of Commerce imposed export controls over security concerns. For about three weeks, Fable 5 sat frozen while free alternatives like GLM kept shipping and gaining ground. On June 30, the Department lifted the restriction, and Anthropic restored global access on July 1 after tightening its security measures. The catch: multiple testers report it came back noticeably more conservative than its original release.

7. Google DeepMind Drops Nano Banana 2 Light and Gemini Omni Flash Together

Google DeepMind released two connected models on the same day. Nano Banana 2 Light, an image model, can generate a full image in about two seconds — in DeepMind’s own demo, it produced 14 images in 30 seconds versus just two from the standard Nano Banana 2. Gemini Omni Flash, the video half, edits existing footage conversationally: typed instructions can lift on-screen text into floating 3D letters or composite entirely new scenes into a real video clip. An API chains the two models together, letting you go from a single prompt to a fully animated, edited video clip in one continuous workflow.

8. Anthropic Launches Claude Science for Researchers

Anthropic introduced Claude Science, an AI research assistant connected to more than 60 scientific databases including PubMed and bioRxiv. It runs inside a live notebook that writes and executes its own code, can render and measure 3D protein structures, and can send heavy computational jobs — like protein-structure prediction — straight to a cloud compute cluster after a single approval. It can also assign a team of agents to read across thousands of papers, synthesize findings, and produce a formatted literature review, complete with a dedicated reviewer agent that checks citations for accuracy.

9. Google’s NotebookLM Turns Study Notes Into Short Vertical Videos

NotebookLM added a “Video Overview” feature that converts uploaded notes, PDFs, or links into short, narrated, illustrated vertical videos — reel-style clips built for studying a specific topic you point it toward. The visuals run on Nano Banana 2 Lite, fast enough to render an entire short animated clip in seconds.

10. Gamma Moves Directly Inside ChatGPT

Gamma, the AI tool that builds full presentations, documents, and websites from a single prompt, is now accessible from inside ChatGPT’s apps menu. Users can type a prompt describing the deck they want, and ChatGPT hands it off to Gamma, which generates a complete, branded slide deck — title slide, custom visuals, and themed layouts — that can then be fine-tuned inside Gamma itself.

11. xAI Launches a No-Code Voice Agent Builder on Grok

xAI released Voice Agent Builder, letting anyone create a phone-answering AI agent in about two minutes without writing code. Users describe the agent’s job, choose or clone a voice, and connect it to their own documents so it answers from real business information — and it can take real actions, like issuing a refund on a live call. It supports 25 languages, with pricing starting around 5 cents per minute.

12. Replit Launches a Native Desktop App

Replit, the browser-based app-building tool, is now available as a native desktop app for Mac and Windows. The upgrade allows multiple projects to run in separate workspaces simultaneously — one building a slide deck, another running a data dashboard — without juggling browser tabs, while keeping Replit’s existing security center and dependency tools intact.

Bonus Round-Up: 6 More AI Updates This Week Worth Watching

OpenAI’s Codex Macro Pad: A dedicated physical control pad with programmable buttons for AI coding shortcuts, slated to launch July 15.
ComfyUI’s Comfy MCP: Lets AI models like Claude control an entire image/video generation pipeline conversationally, including pulling designs from Figma or scripts from Notion.
Notion’s HTML Block: Turns any plain page into an interactive timeline, dashboard, prototype, or quiz using AI, without leaving Notion.
Google AI Studio Design Variations: Generates multiple visual styles for an app instantly, so you can pick a look with one click.
Google AI Studio Shared Chat History: Lets creators share the entire prompt trail behind a built app, not just the finished product.
OpenClaw Mobile App: The free autonomous AI agent tool now has a dedicated iPhone and Android app instead of running only through third-party chat platforms.

What These AI Updates This Week Mean for You

Put together, this week’s AI updates this week point to one clear pattern: no single AI model is safe from being outpaced, undercut, or restricted overnight. Free, open-source models out of China are closing the performance gap on premium U.S. tools within weeks, not years. At the same time, geopolitics — not just competition — can take a flagship model offline with almost no warning, as Claude Fable 5’s export-control episode showed. The smartest move for builders and everyday users alike isn’t loyalty to one AI brand; it’s flexibility — tools like Hermes and Zcode that let you switch models on the fly are becoming just as important as the models themselves.

Frequently Asked Questions

What is the biggest AI update this week?
The return of Claude Fable 5 after its U.S. export-control suspension, alongside a free tool that merges ChatGPT and Claude into a single, higher-scoring AI response.

Is GLM 5.2 really free?
Yes. GLM 5.2 from Z.ai is free and open-source, and its new Zcode app includes a generous daily free token allowance before paid tiers kick in.

Why did Meta reportedly ban Claude Code?
Reported concerns center on “distillation” — the risk that Meta’s own AI could inadvertently train on code quality from a rival’s model, creating potential intellectual property exposure.

AI Updates This Week: 12 Explosive Moves That Just Rewrote the ChatGPT vs Claude War

Must read