gemini-4IMAGE

Gemini 4 Argon passes all four KernelBench-CUDA kernels, beating Gemini 3.8 Flash on each

Argon's four hand-written CUDA kernels all passed final checks on an RTX PRO 6000, while Gemini 3.8 Flash passed three of four, with Argon up to 5x faster on Native Sparse Attention.

Elliot Arledge

@elliotarledgeArchived 21 days agoOriginally posted 2026.10.14

Gemini 4 Argon passes all four KernelBench-CUDA kernels, beating Gemini 3.8 Flash on each

Elliot Arledge @elliotarledgeOriginally posted 2026.10.14

Open original on X

Gemini 4 Argon [high] on KernelBench-CUDA: All four hand-written CUDA kernels passed their final checks on RTX PRO 6000: 6th, 4th, 8th and 9th. Gemini 3.8 Flash passed three of the four, and Argon beats it on every one, Native Sparse Attention by 5x. One problem per reply below.

x.com/elliotarledge/status/2110370049961918708Archived 2026.10.15 11:34