Method Drift›Mixture-of-experts routing
Superseded baseline#48 of 1,370 most-superseded
GRPO
Mixture-of-experts routing
superseded — cited as a baseline and beaten by newer methods
1 papers critique it · 2 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites GRPO as a baseline.
“GRPO can suffer from abrupt reward collapse on MoE models.”
— Towards Stable and Effective Reinforcement Learning for Mixture-of-Experts
Beaten on benchmarks
Head-to-head results where a newer method reports beating GRPO. Values are copied from the source paper's tables — verify against the cited paper.
- Towards Stable and Effective Reinforcement Learning for Mixture-of-Experts
GMPO+RS (RSPO) beats GRPO · Pass@1 [Math]
77.1 vs 71.5
- Towards Stable and Effective Reinforcement Learning for Mixture-of-Experts
GMPO+RS (RSPO) beats GRPO · Pass@1 [Code]
85.2 vs 70.7
- MESA: Improving MoE Safety Alignment via Decentralized Expertise
MESA beats GRPO · GSM8K [Model: Qwen3-30B-A3B]
96.44 vs 96.36
Newer alternatives
Recent methods in the same sub-problem, not yet superseded in the knowledge base.
- PADDPADD: Path-Aligned Decompression Distillation for Non-Router Teacher to Guide MoE Student LearningJun 9, 2026
- May 30, 2026
- May 29, 2026
- May 1, 2026
- Apr 30, 2026
- Feb 9, 2026
- SocialNav-MoESocialNav-MoE: A Mixture-of-Experts Vision Language Model for Socially Compliant Navigation with Reinforcement Fine-TuningDec 15, 2025
- OrdMoEOrdMoE: Preference Alignment via Hierarchical Expert Group Ranking in Multimodal Mixture-of-Experts LLMsNov 24, 2025
- router-aware approach to optimize importance sampling weightsTowards Stable and Effective Reinforcement Learning for Mixture-of-ExpertsOct 27, 2025
- Mix- and MoE-DPOMix- and MoE-DPO: A Variational Inference Approach to Direct Preference OptimizationOct 9, 2025