TraceForge for Distilling Reasoning Traces from Teacher Models
October 20, 2026
TraceForge provides a method to generate high-quality reasoning traces for distillation by using the teacher's context and answer as inputs. This avoids the terms-of-service issues associated with replaying encrypted reasoning blocks from proprietary models.
HOW THIS AFFECTS YOU
●
builderYou can more easily distill reasoning capabilities into smaller, open-source models without violating provider policies.
●
founderYou can improve your proprietary models' reasoning performance by using this structured distillation approach.