Our paper “TokenSqueeze: Performance-Preserving Compression for Reasoning LLMs” got accepted by NeurIPS 25. See you all in San Diego.