“Unlike other model releases, OpenAI is providing custom kernels alongside its open-source model weights to ensure a highly optimized inference stack is available from day one.”