Changelog

You can now filter dataset benchmark leaderboards by the number of model parameters. On any dataset leaderboard, pick a size range and the rankings refresh to that bucket. The top 3 models in each size category are marked with a 🏅.

Every Gradio Space now auto-serves an /agents.md endpoint, a machine-readable API description that AI agents can read and call directly. Point your coding agents (like Claude Code, Codex, or Pi) at it and they figure out how to use the Space without any setup. Read the documentation to learn more.

Kernels repositories provide an easy way to use kernels: precompiled, optimized for your exact hardware and PyTorch version, ready for torch.compile, and yields 1.7–2.5× speed-ups over baseline PyTorch.

You can now browse and load Kernels from the Kernels Hub.

PRO users can now continue using ZeroGPU Spaces above their daily included quota.

Over-quota usage requires purchasing pre-paid credits. The price is $1 per 10 minutes of over-quota ZeroGPU usage.

You can now upload traces from your agents (Claude Code, Codex, Pi) directly to Hugging Face Datasets. The Hub auto-detects trace formats and tags your dataset as Traces, with a dedicated viewer for browsing sessions, turns, tool calls, and model responses.

No preprocessing needed, just upload the JSONL files from your local session directories as-is:

Useful for sharing debugging workflows, benchmarking agent behavior across models, or building training data from real coding sessions.