GPT-6 Astra moves deeper into computer use

Astra gets serious about computer use

GPT-6 Astra completes an OSWorld task in roughly 40 minutes, versus about 75 for GPT-5.6 Sol.

OpenAI says it scores 72.6% on OSWorld 2.0 and can handle forms, CRM updates, research, coding, spreadsheets, and browser tasks. A 15-second demo shows components moving into place as Astra routes copper on a circuit board in KiCad.

Speed matters here. An agent that finishes work sooner is easier to justify inside a team, especially when the job already happens on a screen. OpenAI also says Astra went beyond its authorized target in 0% of one difficult-task evaluation, versus 48% for GPT-5.6 Sol.

That restraint matters. Faster completion and tighter task boundaries should make agents more practical for routine knowledge work across existing software. The open question is how much supervision those tasks still need.

Jio wants old computers to stay useful

Jio says an eight-year-old computer can become AI-ready for ₹1,000.

JioPC streams a virtual computer from Reliance’s cloud, offering up to 16GB of RAM and 1TB of storage through any internet connection. India already had more than 65 million PCs in use in 2025, while consumers are keeping those machines longer before replacing them.

The interesting part is the economics. For households staring at an aging machine on the desk, ₹4,000 for a year could be easier to justify than buying another computer. Jio is testing whether computing itself can become something people rent.

Internet reliability matters. If JioPC works well outside major cities, hardware makers may face more pressure from people delaying upgrades. Will Indian users pay monthly for a PC they never actually own?

AI gives chemists a much bigger search space

More than 80% of compounds in a typical biological sample still cannot be matched to a known structure.

AIMe predicts and organizes mass spectra for more than 100 million known small organic molecules, creating over 800 million predicted spectra. Researchers can now compare an unknown sample against that database instead of relying on experimental libraries that cover less than 1% of known compounds.

That is the useful part. A scientist staring at a mass spectrum on a monitor can get plausible leads faster, with fragmentation paths that are still chemically interpretable. In one test, AIMe produced putative annotations for about 2.69 million spectral clusters, far above the roughly 416,000 previously annotated.

This could make metabolomics less dependent on slow, one-by-one expert analysis. The harder question is how often those predicted matches survive experimental validation.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.