NanoGPT Speedrun Frontier

The NanoGPT Speedrun Frontier tracks how quickly different AI models can close the gap toward a human performance benchmark, with Fable 5 leading at 81.7% of the gap closed in 8.7 days using the claude-code harness. The leaderboard shows performance metrics across 20 models, measuring success based on test scores, token usage, and time-to-result, with faster models like Grok 4.6 and Muse Spark 1.2 achieving meaningful progress in under a day.