Qwen2.5 32B DialogueReason
DialogueReason is StepFun's post-trained derivative of Qwen2.5-32B-Base. It uses rule-based reinforcement learning and Open-Reasoner-Zero data to develop multi-turn reasoning between simulated experts.
Research release
Generate a discussion before the answer
The example prompt asks the model to produce an expert discussion inside play tags, then a final result inside answer tags. This is a generated dialogue within one model, rather than evidence of independent external experts.StepFun Qwen2.5-32B-DialogueReason model card
Judge the answer after the dialogue
Practical guidance. A long discussion can look persuasive while sharing one mistaken assumption. Score the final answer against evidence or an independent checker. Compare shorter reasoning prompts to see whether the extra dialogue helps your task.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | Qwen2.5-32B-DialogueReason |
| Base model | Qwen2.5-32B-Base, developed by Alibaba's Qwen team.StepFun Qwen2.5-32B-DialogueReason model card |
| License | Apache 2.0 is declared for this release.StepFun Qwen2.5-32B-DialogueReason model card |
| Training contribution | Rule-based reinforcement learning for dialogue reasoning.StepFun Qwen2.5-32B-DialogueReason model card |
Related models
PaCoRe 8B is a reasoning model trained to combine parallel solution paths.
Launch date not confirmedResearch releaseDownloadable- RLVR 8B 0926
StepFun · PaCoRe
A Qwen3-8B-Base derivative with reasoning-oriented post-training, used as the initial PaCoRe checkpoint.
Launch date not confirmedResearch releaseDownloadable