StepFunPaCoRe
RLVR 8B 0926
RLVR-8B-0926 is the initial checkpoint in StepFun's PaCoRe study. It applies reasoning-oriented post-training to Qwen3-8B-Base and remains a separate downloadable model.
Research release
Keep starting and final checkpoints distinct
The model card describes the wider PaCoRe project, including final-model results. Those results should not be attributed to this starting checkpoint without a matching experiment.StepFun RLVR-8B-0926 model card
Use it to study the effect of later training
Practical guidance. This release is useful for a controlled comparison with PaCoRe. Keep the prompt, sampling settings and computation budget fixed where possible. Otherwise changes in inference can obscure changes caused by training.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | RLVR-8B-0926 |
| Base model | Qwen/Qwen3-8B-Base.StepFun RLVR-8B-0926 model card |
| License | MIT.StepFun RLVR-8B-0926 model card |
| Training stage | Initial research checkpoint; PaCoRe-8B is the project's final coordinated-reasoning release.StepFun RLVR-8B-0926 model card |
Related models
PaCoRe 8B is a reasoning model trained to combine parallel solution paths.
Launch date not confirmedResearch releaseDownloadable