Fact◆Article
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.
Fable wrote the first megakernel submitted to KernelBench-Mega, achieving an 18.71x speedup with a single kernel launch per token, versus 4-14 launches for competing models, signaling rapid AI progress on tasks central to automating AI R&D itself. ✦ AI generated
Jack Clark · Import AI · 2026-07-06 · original ↗
This solution is particularly impressive because “torch.profiler shows exactly ONE cooperative kernel launch per decoded token”. By comparison, every other high-scoring entry decomposed the problem into anywhere from 4 to 14 separate kernel launches per token.
Read full article ↗excerpt · fair-use quotation
- ·Autonomously authored first megakernel on KernelBench-Mega
- ·Achieves 18.71x speedup over optimized PyTorch baseline
- ·Uses one cooperative kernel launch per decoded token
- ·Beats every other frontier model's multi-kernel approach
- ·Fable: single kernel launch per token
- ·Rivals: 4 to 14 separate launches per token
- ·Fewer launches signals more efficient AI-authored code
- ·Beat every other frontier model's approach
- ·Signals fast progress on automating AI R&D
- ·Task: writing low-level GPU code, not just app code
Around this claim