This one dropped today and the reaction online has been immediate. Anthropic just shipped computer use inside Claude Code, meaning Claude can now open your apps, click through your UI, take screenshots, and test what it built, all without leaving the command line. It is available now in research preview on Pro and Max plans on macOS, and you enable it with a single command.

The capability itself is worth sitting with for a moment, because it represents a genuinely different kind of tool. Previously, Claude Code would write your code and hand it back to you. You would run it, find the bugs, describe them, and feed that back in. That loop is now closed. In a single prompt, Claude can write the code, compile it, launch the app, click through the interface, find what broke, fix it, and verify the fix. No handoff. No switching tabs. No copy-pasting error messages back and forth.

The demo floating around today shows Claude opening a pixel art app called SPM3000, hitting a buffer overrun error mid-render, reading the stack trace, locating the bug in the source, patching it, and confirming the fix, all autonomously, all from the terminal. The response from developers has been somewhere between impressed and slightly unsettled, which is usually a reliable sign that something significant just shipped.

The way people are talking about this is that it collapses the last remaining gap between writing software and owning its quality. Claude was already writing code most developers would be happy to ship. Now it is also doing the part developers like least.

Write it. Run it. Break it. Fix it. Ship it. Claude Code just took the whole loop.

Alibaba's Qwen team dropped Qwen 3.6 Plus Preview on OpenRouter today, and it is not a minor release. This is their flagship model, a meaningful step up from the 3.5 series with stronger reasoning, more reliable agentic performance, better coding and front-end development, and a one million token context window. In benchmarks it sits at or above the leading frontier models. This is Alibaba swinging at the top of the market.

To put the context window in perspective, a million tokens is roughly the equivalent of ten full-length novels worth of text that the model can read and reason over in a single session. Most people will never need that much. But the fact that it is available, for free, from a Chinese lab, while Western competitors charge significant premiums for smaller windows, says something about where this market is heading.

The numbers are hard to ignore. Claude Opus 4.6 charges up to $25 per million tokens for a fraction of the context window. Qwen 3.6 Plus Preview offers five times that window for nothing.The trade-off is that during the free period, your prompts and completions will be collected and used to improve the model. You are not getting something for nothing so much as paying with data instead of dollars. Whether that feels reasonable depends on what you are building.

The bigger picture is what this says about where AI is heading. A year ago, a one million token context window was considered ambitious. Today it is being given away for free to drive adoption. The pressure on paid APIs is real, and it is coming from open source and Chinese labs simultaneously. OpenAI and Anthropic have been able to charge premium prices because they have been premium products. That gap is closing faster than most people expected, and releases like this one are exactly why.

Alibaba shipped a flagship. The free part is just the attention-grabber.

Buried inside Claude's code this week, researchers found references to two unannounced modes sitting alongside Chat, Cowork, and Code: one called Operon, one called Epitaxy. No official announcement. No docs. Just enough context to piece together what they are.

Operon appears to be built specifically for biological research. CRISPR, if you are not familiar, is a gene-editing technology that lets scientists precisely cut and modify DNA and is one of the most consequential tools in modern medicine. The early interface shows prompts designed around real biology workflows, suggesting something closer to a research collaborator than a chatbot. Epitaxy appears to be aimed at materials science, the field that engineers new physical materials from semiconductors to superconductors, with similarly domain-specific tooling underneath.

The strategic read is straightforward. Anthropic is not just building a smarter general assistant. They are building Claude vertically, one scientific field at a time.

Everyone else is still arguing about system prompts. Anthropic is shipping modes for CRISPR screens.