HOT RADAR
agentHEAT 7.3°

AI CURATED HOT STORY · 10/30/2026

ModelBest Open-Sources ForgeKernel: AI Agent Beats DeepSeek's Hand-Written Operators by 9.8%

1 reports archived1 independent sourcesupdated 10/30/2026, 19:05:06
Synthesis & Latest Updates
1 Sources Cross-Validated

ModelBest has open-sourced 'ForgeKernel', an autonomous agent that generated a FlashMLA Prefill operator outperforming DeepSeek's hand-written version by 9.8% on NVIDIA B200 after 22.5 hours of self-evolution without human code. This milestone demonstrates AI's capability to autonomously optimize low-level infrastructure, signaling a paradigm shift from manual coding to AI-generated solutions with human architectural oversight.

LATEST/ModelBest has open-sourced 'ForgeKernel', an autonomous agent that generated a FlashMLA Prefill operator outperforming DeepSeek's hand-written version by 9.8% on NVIDIA B200 after 22.5 hours of self-evolution without human code. This milestone demonstrates AI's capability to autonomously optimize low-level infrastructure, signaling a paradigm shift from manual coding to AI-generated solutions with human architectural oversight.

Hourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first
  1. 量子位官网T2·88 pts
    • ForgeKernel achieved a 9.8% performance gain over FlashMLA operators in 22.5 hours via 42 iterations without source code reference.
    • The project is now open-source, supporting both domestic (Ascend) and NVIDIA hardware, proving AI can break through software ecosystem barriers.
    • R&D paradigm shift: Human roles evolve from code craftsmen to system architects who define goals, environments, and validation standards for AI.
ModelBest Open-Sources ForgeKernel: AI Agent Beats DeepSeek's Hand-Written Operators by 9.8% | Today for AI