---
type: "post"
id: "researchers-at-shanghai-ai-laboratory-and-monash-university-7b5fc1"
slug: "researchers-at-shanghai-ai-laboratory-and-monash-university-7b5fc1"
title: "Researchers at Shanghai AI Laboratory and Monash University have introduced RePro , a process-level reward for training reasoning models to produce shorter, more accurate solutions."
description: "**Thesis:** RePro makes the reasoning process itself an optimization target by turning process quality into a reward used during post-training."
retrieval_nugget: "**Thesis:** RePro makes the reasoning process itself an optimization target by turning process quality into a reward used during post-training."
published_at: "2026-10-15"
updated_at: "2026-10-15"
record_date: "2026-10-15"
date_kind: "published_at"
topics: ["models"]
source_urls: ["https://arxiv.org/abs/2512.01925"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:researchers-at-shanghai-ai-laboratory-and-monash-university-7b5fc1"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:4ae69cd5008d1c521988c45b192c25a060940f4cb7684f803a5906c8345ac388","reviewed_at":"2026-10-15T13:04:54.915Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":1,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"RePro makes the reasoning process itself an optimization target by turning process quality into a reward used during post-training.","observed_facts":[{"text":"The abstract says RePro uses dual scores for reasoning-process intensity and stability, aggregated into a process-level reward.","source_urls":["https://arxiv.org/abs/2512.01925"]},{"text":"The abstract says that reward is integrated into reinforcement learning with verifiable rewards and reports experiments across multiple RL algorithms and diverse LLMs on mathematics, science, and coding benchmarks.","source_urls":["https://arxiv.org/abs/2512.01925"]},{"text":"The arXiv record lists submission on 1 December 2025, revision on 7 April 2026, and acceptance by ICLR 2026.","source_urls":["https://arxiv.org/abs/2512.01925"]}],"mechanism":"RePro scores reasoning-process intensity and stability, combines the scores into a process-level reward, and integrates that reward into reinforcement learning with verifiable rewards (RLVR).","why_now":"The paper identifies overthinking and excessively long reasoning chains as behaviors that can impair performance, making process-level supervision a relevant post-training question.","implications":["If this signal generalizes, teams training reasoning models may be able to optimize for useful progress through a solution rather than rewarding length or relying only on final-answer correctness."],"evidence_boundary":"The supplied abstract describes the scoring design and evaluation scope but gives no numerical results, ablations, or benchmark-by-benchmark comparisons; it does not establish the size or robustness of any gains.","watch_conditions":["Look for the full paper’s quantitative comparisons and ablations showing whether intensity and stability each contribute, and whether shorter reasoning preserves accuracy across tasks."],"related_records":[],"new_branch_reason":"This is a distinct analytical record because no prior New Runtime record was supplied for RePro’s process-level reward mechanism."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/researchers-at-shanghai-ai-laboratory-and-monash-university-7b5fc1/","markdown":"https://newruntime.com/posts/researchers-at-shanghai-ai-laboratory-and-monash-university-7b5fc1.md","json":"https://newruntime.com/posts/researchers-at-shanghai-ai-laboratory-and-monash-university-7b5fc1.json"}
---

# Researchers at Shanghai AI Laboratory and Monash University have introduced RePro , a process-level reward for training reasoning models to produce shorter, more accurate solutions.

## Retrieval answer

**Thesis:** RePro makes the reasoning process itself an optimization target by turning process quality into a reward used during post-training.

**Thesis:** RePro makes the reasoning process itself an optimization target by turning process quality into a reward used during post-training.

The paper frames chain-of-thought as a sequence of updates toward resolving a problem. **Mechanism:** RePro scores reasoning-process intensity and stability, combines the scores into a process-level reward, and integrates that reward into reinforcement learning with verifiable rewards (RLVR). This is a training signal about the path to an answer, not simply a rule that asks a model to be concise.

**Why now:** The paper identifies overthinking and excessively long reasoning chains as behaviors that can impair performance, making process-level supervision a relevant post-training question. The arXiv record says the paper was submitted on 1 December 2025, revised on 7 April 2026, and accepted by ICLR 2026.

**First implication:** If this signal generalizes, teams training reasoning models may be able to optimize for useful progress through a solution rather than rewarding length or relying only on final-answer correctness. The paper reports experiments using multiple RL algorithms and diverse LLMs, with mathematics, science, and coding benchmarks.

**Evidence boundary:** The supplied abstract describes the scoring design and evaluation scope but gives no numerical results, ablations, or benchmark-by-benchmark comparisons; it does not establish the size or robustness of any gains.

**First watch condition:** Look for the full paper’s quantitative comparisons and ablations showing whether intensity and stability each contribute, and whether shorter reasoning preserves accuracy across tasks.

This is a distinct analytical record because no prior New Runtime record was supplied for RePro’s process-level reward mechanism. Source: https://arxiv.org/abs/2512.01925

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Models](https://newruntime.com/topics/models.md): Explore the models topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)