Sakana AIRLT
RLT-32B
The RLT-32B student adapts Qwen2.5-32B-Instruct with supervised reasoning traces produced by a 7B teacher.
Research release
The student uses supervised fine-tuning
Reinforcement learning optimizes the teacher to produce useful explanations for students. The released student learns from those traces through supervised fine-tuning. Its name does not mean that the student itself underwent that reinforcement-learning stage.RLT-32B official model card
Working with this release
Practical guidance. Keep teacher and student sizes explicit when comparing training methods. Check final answers and intermediate reasoning independently.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | SakanaAI/RLT-32B |
| Student ancestry | Qwen2.5-32B-InstructRLT-32B official model card |
| Teacher | A 7B reinforcement-learned teacher, including for the 32B student.RLT-32B official model card |
| Terms | Apache 2.0; the card describes research-only intent and discourages commercial or critical deployment.RLT-32B official model card |
Related models
A 7B reasoning student trained from a reinforcement-learned teacher.
Launch date not confirmedResearch releaseDownloadable