---
type: "post"
id: "tested-on-five-7b-mlrms-with-consistent-gains-fd8b39"
slug: "tested-on-five-7b-mlrms-with-consistent-gains-fd8b39"
title: "Tested on five 7B MLRMs with consistent gains on CHAIR, POPE, AMBER, and GPT-4 benchmarks: paper ."
description: "Hao Fang, Jinyu Li, Jiawei Kong, Tianqu Zhuang, Kuofeng Gao, Bin Chen, and Shu-Tao Xia published a revised arXiv paper, “Seeing Through the Chain: Understanding and Mitigating Hallucinations in Multimodal Large Reasoning Models.” The current version is v3, revised on October 23, 2026; the first"
retrieval_nugget: "Hao Fang, Jinyu Li, Jiawei Kong, Tianqu Zhuang, Kuofeng Gao, Bin Chen, and Shu-Tao Xia published a revised arXiv paper, “Seeing Through the Chain: Understanding and Mitigating Hallucinations in Multimodal Large Reasoning Models.” The current version is v3, revised on October 23, 2026; the first"
published_at: "2026-10-27"
updated_at: "2026-10-27"
record_date: "2026-10-27"
date_kind: "published_at"
topics: ["models"]
entities: ["C3PO","arXiv"]
related_patterns: ["scientific-agents-need-artifact-evidence"]
novelty_label: "new_fact"
what_changed: "A new paper introduces C3PO to reduce hallucinations in multimodal reasoning models, reporting gains across five 7B models and multiple benchmarks; the prior benchmark story concerns scientific-paper retrieval."
prior_refs: ["memory:1826fd59-0f0a-8f9c-a452-ee8a03359de6"]
source_url: "https://arxiv.org/abs/2602.03380"
source_urls: ["https://arxiv.org/abs/2602.03380"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:tested-on-five-7b-mlrms-with-consistent-gains-fd8b39"
status: "published"
visuals: []
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/tested-on-five-7b-mlrms-with-consistent-gains-fd8b39/","markdown":"https://newruntime.com/posts/tested-on-five-7b-mlrms-with-consistent-gains-fd8b39.md","json":"https://newruntime.com/posts/tested-on-five-7b-mlrms-with-consistent-gains-fd8b39.json"}
---

# Tested on five 7B MLRMs with consistent gains on CHAIR, POPE, AMBER, and GPT-4 benchmarks: paper .

## Retrieval answer

Hao Fang, Jinyu Li, Jiawei Kong, Tianqu Zhuang, Kuofeng Gao, Bin Chen, and Shu-Tao Xia published a revised arXiv paper, “Seeing Through the Chain: Understanding and Mitigating Hallucinations in Multimodal Large Reasoning Models.” The current version is v3, revised on October 23, 2026; the first

Hao Fang, Jinyu Li, Jiawei Kong, Tianqu Zhuang, Kuofeng Gao, Bin Chen, and Shu-Tao Xia published a revised arXiv paper, “Seeing Through the Chain: Understanding and Mitigating Hallucinations in Multimodal Large Reasoning Models.” The current version is v3, revised on October 23, 2026; the first version was submitted on February 3, 2026.

The authors introduce C3PO, a training-based framework combining chain-of-thought compression with contrastive preference optimization. Their stated premise is that reasoning mechanisms can increase reliance on language priors over visual inputs, producing traces with fewer visual cues and redundant text.

C3PO filters selected reasoning tokens, uses preference pairs built with AI feedback, and adds hallucination-inducing prompts to create negative training signals. The paper reports consistent reductions across multimodal reasoning models and evaluations including CHAIR, POPE, AMBER, and a GPT-4-based benchmark.

The source does not establish independent replication, production performance, the cost of training or inference, or whether the method improves reliability on real user workloads.

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)