PyTorch Adds 10 New AI Tools That Save Time and Money

Think of PyTorch as the world's favorite free Lego set for building artificial intelligence. Thousands of developers use it to create everything from photo filters to self-driving car brains. Today, the PyTorch community officially welcomed 10 new add-on tools that make AI easier, cheaper, and more powerful.

One standout, called Perforated, helps AI learn from far fewer examples. Normally, AI needs tons of labeled data—pictures, text, or sensor readings that humans have sorted by hand. Perforated tweaks the learning process so AI squeezes more value out of what it already has. The makers say teams using it reduced AI errors by 70% with existing data, saving days or weeks of tedious data collection.

Another tool, AReaL, helps AI connect to real-world apps and tools—like letting a chatbot actually book a flight or check your calendar. It acts as a bridge between the AI's brain and outside systems, making "agentic AI" easier to build. TorchJD tackles a different problem: training AI that must balance multiple goals at once, like a delivery robot that wants speed but also safety. And RLinf lets AI practice through trial and error inside simulated worlds, which is how robots learn to walk without breaking anything.

The big picture? These new tools lower the barrier for building capable AI. You don't need a massive data center or a team of PhDs. That means smaller companies could build better AI products, and everyday apps might get smarter sooner. The catch is that most of these tools are early-stage, so they'll need time to grow—but the foundation just got a whole lot stronger.