[JyrkiAlakuijala]

JPEG XL doesn't use the kind of rANS that Microsoft has patented.

JPEG XL decides the codes at encoding time and does context modeling the same way as WebP lossless and Brotli, by deciding which entropy codes to use explicitly.

Microsoft's rANS patent is supposedly centralized around updating rANS codes at decoding time (based on past symbols). This is slightly more efficient for density, but much slower and may negate the speed benefits that rANS brings. For practical implementations JPEG XL/Brotli way is quite a bit better.