A new AI model named Fable has demonstrated the ability to write efficient GPU kernels, achieving an 18.71X speedup on an Nvidia RTX Pro 6000 Blackwell Workstation Edition compared to an optimized PyTorch baseline. This performance surpasses other leading models like Claude-Opus-4.8, GLM-5.2, and GPT-5.5 on the KernelBench-Mega benchmark, indicating advancements in AI's capacity for R&D automation. Separately, research from the Center for AI Safety and Scale Labs shows a significant increase in AI's success rate for automating online freelance projects, rising from 2.5% to 16.1% between October 2025 and July 2026, with Fable achieving the highest success rate among tested frontier models. AI
IMPACT Sets new SOTA on GPU kernel optimization and signals accelerating AI capability in automating complex online work tasks.
RANK_REASON AI model demonstrates superior performance on a benchmark and research shows increased AI capability in automating freelance tasks.
Read on Import AI (Jack Clark) →
- Center for AI Safety
- Claude-Opus-4.8
- fable
- GLM-5.2
- GPT-5.5
- KernelBench-Mega
- Nvidia RTX Pro 6000 Blackwell Workstation Edition
- PyTorch
- Remote Labor Index
- Scale Labs
- Triton
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
