Rohan Paul (@rohanpaul_ai)T2·78 pts
- Reasoning traces can be compiled into parameterized Python functions reused on CPU instead of re-running the LLM
- Matched or beat the source model on 7 of 8 benchmarks, scoring 93.3% vs 83.3% on AIME '25
- Covered at least 90% of rephrased or re-numbered variants at 16x to 291x lower cost