Every run starts with a question. Can we make this model better? You've now asked it 1,000,000,000 times. (how many failed? No comment 😂) Thank you doesn't cover it, but thank you.

You made models better a billion times. We spent the year making runs better. Pinned baselines, semantic coloring, display names, in-app diffs. An iOS app for loss curves anywhere. W&B LEET for the terminal faithful. An MCP server so your coding agent can read runs too.

W&B Weave got rebuilt for production agents. It traces sessions the way agents actually execute them and catches regressions before users do. Registry now gives models, datasets, prompts, and more the versioning and governance platform teams keep asking for.

The loop got shorter too. @CoreWeave ARIA reads your experiments, forms a hypothesis, launches the next run, and scores it against your baseline. Serverless Inference, Serverless RL, and Serverless SFT handle the compute underneath, so you never think about infrastructure.

None of this is a finish line. AI development runs in a loop now. You ship, you watch production, you learn, you retrain, you ship again. A billion runs taught us what that loop needs, and we're building every piece of it.

So THANK YOU!!! To everyone who ever logged a run and then tried one more config after swearing that was the last one. The first billion runs were yours. So is the next billion. Full story, and everything we've been shipping, below.

Here's to 1 billion more!