# ForgeKernel, an AI-written operator, tops two kernel benchmarks after DeepSeek's DeepInfer release

> Original title: DeepSeek 开源看家算子的同一周，AI 超过了它

- Source: Leiphone
- Published: 2026-10-26T06:58:00.000Z
- Bad Influence: https://badinfluence.news/items/kjz3upnxce1lo7gbpt3ljqi8r
- Original: https://www.leiphone.com/category/industrynews/IaMw7vRrTCadlbCY.html

## Summary

DeepSeek recently open-sourced its DeepInfer inference framework, following its model weights and operator libraries such as FlashMLA and DeepGEMM. In the same week, ForgeKernel, an operator written entirely by AI with no manual line-by-line tuning, ranked first on NVIDIA's SOL-ExecBench FlashInfer-Bench collection and on the Ascend CANNBench A3 hardware leaderboard as of October 22, with a 1.835× benchmark speedup, 25% ahead of second place.

## Why it matters

The report pairs DeepSeek's open-sourced DeepInfer framework with AI-written kernels that top two operator benchmarks, showing how kernel efficiency sets GPU speed and token price.