Data◆Article
Fable's AI-written megakernel is the fastest and most efficient solution ever submitted to KernelBench-Mega, using a single cooperative kernel launch per token where every other high-scoring entry needed 4 to 14 — a sign AI is closing in on the R&D tasks that underlie recursive self-improvement.
Fable's Cuda-written megakernel hit an 18.71X speedup over an optimized PyTorch baseline, beating Claude Opus 4.8, GLM-5.2, and GPT-5.5 on the same benchmark, and did so with a single kernel launch per token versus competitors' 4-14. ✦ AI generated
Jack Clark (Import AI) · Import AI · 2026-07-06 · original ↗
Fable achieved an 18.71X speedup by writing Cuda code on an RTX PRO 6000 Blackwell, compared against an optimized PyTorch baseline. For calibration, other attempts at this get 14.4X (Claude Opus 4.8, writing Triton), 11.14X (GLM-5.2, Triton), and 4.34X (GPT 5.5, Triton).
Read full article ↗excerpt · fair-use quotation
- ·Fable hit 18.71X speedup over optimized PyTorch baseline
- ·Beat Claude Opus 4.8, GLM-5.2, and GPT-5.5
- ·Used single cooperative kernel launch per token
- ·Rivals needed 4 to 14 launches per token
- ·Efficiency gain signals AI nearing self-improvement R&D tasks
- ·Fable wrote raw Cuda; rivals used Triton
- ·Single-launch design shows deeper hardware-level optimization
Around this claim
This moment responds to
supports → Fable's AI system autonomously wrote a CUDA megakernel that achieves an 18.71x speedup over an optimized PyTorch baseline using a single cooperative kernel launch per token, beating every other frontier model's multi-kernel approach on KernelBench-Mega.Jack Clark · Import AIextends → Fable's AI system autonomously wrote a CUDA megakernel that achieves an 18.71x speedup over an optimized PyTorch baseline using a single cooperative kernel launch per token, beating every other frontier model's multi-kernel approach on KernelBench-Mega.Jack Clark · Import AI