Have you by any chance tried this with GPT 4.1 too (also 1M context)?