Understand this story
On October 28, 2026, ModelBest's OpenBMB reposted a message from Yuxuan Li about the ForgeKernel kernel optimization framework, saying it looks forward to the forthcoming paper and more results from the Forge series. The framework belongs to the Forge series and is said to rank first in tests on NVIDIA SOL-ExecBench (FlashInfer-Bench) and Huawei CANN-Bench. On October 30, ModelBest announced the open-sourcing of ForgeKernel; its agent completed 42 rounds of iteration in about 22 hours, producing 32 kernel versions, and achieved 1.098x performance on NVIDIA B200 for the DeepSeek V4.1 FlashMLA Prefill implementation, at 472 microseconds versus a baseline of 518 microseconds. The open-source release includes the optimized fused prefill kernel, the kernels covered in the paper spanning attention, Gated DeltaNet, Kimi Delta Attention, GEMM and convolution, as well as the ForgeKernel harness that can be used to optimize one's own kernels.
AI-generated from public reports
Latest development · Report
OpenBMB Open-Sources ForgeKernel: DeepSeek V4.1 Prefill Kernel 9.8% FasterDevelopments
2 confirmed developments
- Source:English text is being prepared.
OpenBMB says it looks forward to the Forge series papers and more results
Source:English text is being prepared.
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 30
OpenBMB
OpenBMB Open-Sources ForgeKernel: DeepSeek V4.1 Prefill Kernel 9.8% Faster
OpenBMB announced the open-sourcing of ForgeKernel; its agent completed 42 iterations in about 22 hours and produced 32 kernel versions, ultimately achieving 1.098x performance on the DeepSeek V4.1 FlashMLA Prefill on an NVIDIA B200, at 472 µs versus the 518 µs baseline. The official FlashMLA implementation remained closed throughout; the agent could only run and time its black-box CLI and could not read source code, documentation, or profiling data. The open-sourced content includes the optimized DeepSeek V4.1 fused prefill kernel, the kernels from the paper covering attention, Gated DeltaNet, Kimi Delta Attention, GEMM, and convolution, as well as the ForgeKernel harness that can be used to optimize one's own kernels; across 27 operators, ForgeKernel matched or outperformed 18 of 22 expert-library kernels and outperformed all 7 kernels published by three kernel-generation systems.
Oct 29
ModelBest OpenBMB
ModelBest's ForgeKernel Tops NVIDIA SOL Bench FIB Track
ModelBest OpenBMB announced that ForgeKernel ranks first on both the FIB track of NVIDIA SOL Bench and Huawei CANNBench. Currently, ForgeKernel's goal is to surpass top human experts on one of DeepSeek's most complex operators.
Oct 28
ModelBest OpenBMB
ModelBest focuses on the ForgeKernel kernel optimization framework
Yuxuan Li announced that, based on previous findings from JustGoal, they redesigned their approach to kernel optimization and launched a self-evolving kernel optimization framework called ForgeKernel (part of the Forge series). It ranked first on both NVIDIA's SOL-ExecBench (FlashInfer-Bench) and Huawei's CANN-Bench, and the paper will be released soon. ModelBest OpenBMB reposted it and said they look forward to the upcoming paper and more results from the Forge series.