Is DPO (Direct Preference Optimization) superseded?

Q: Is DPO (Direct Preference Optimization) superseded?

DPO (Direct Preference Optimization) (LLM reasoning / chain-of-thought): cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here. 1 paper(s) critique it, 0 beat it on benchmarks — #262 of 772 most-superseded. Sub-problem: cluster led by VL-Rethinker-7B. Newer alternatives in the same sub-problem include TVI-CoT.

LLM reasoning / chain-of-thought

DPO (Direct Preference Optimization)

Superseded baseline#262 of 772 most-superseded

Cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here

1 papers critique it · 0 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites DPO (Direct Preference Optimization) as a baseline.

While alternatives like Direct Preference Optimization (DPO) simplify the process by using static preference data, their efficacy is limited in tasks requiring dynamic interaction.

— Empowering Lightweight MLLMs with Reasoning via Long CoT SFT

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.

TVI-CoTTVI-CoT: Text-Visual Interleaved Chain-of-Thought Reasoning for Multimodal Understanding
Jun 7, 2026