CompassPlay is introduced, a self-play method that rewards the proposer through gradient alignment that draws on a first-order approximation to learning progress and scores each eligible task without additional solver training.
Abstract
In self-play, a proposer generates verifiable tasks to train a solver. Proposer rewards often depend on the solver's success rate, but equally difficult tasks can differ in their training value. We introduce CompassPlay, a self-play method that rewards the proposer through gradient alignment. The reward favors tasks whose solver loss gradients align with those of reference tasks representing the target capabilities. It draws on a first-order approximation to learning progress and scores each eligible task without additional solver training. Our experiments show gains in performance and training efficiency. In coding self-play with Qwen2.5-Coder-7B, a small external reference set guides task generation. CompassPlay improves average accuracy over AZR's difficulty reward by 1.5 percentage points on in-domain coding and 2.7 on out-of-domain mathematics. In Lean4 theorem proving, CompassPlay matches the difficulty baseline's 150-iteration cumulative coverage with 40\% fewer GPU-hours.
The reason-over-search setting can supply a suitable reward for RLVR on small models, but small-model RLVR needs its own reward-design study rather than a scaled-down copy of a large-model recipe.
Gaurisankar Jayadas, A. Plaat, Álvaro Serra-Gómez et al.· arXiv.org· 0 citations
RLTL;DR is introduced, a promising compacted training paradigm of the form"on this sort of task, keep this sort of thing in mind", which it is hoped will inspire future research on.
Michael Kirchhof, Eleonora Gualdoni, Andrew Szot et al.· arXiv.org· 1 citation
T1, a Mixture-of-Experts model of 122B total trained with reinforcement learning, operating a real shell in a cloud sandbox for up to 300+ tool-call turns per task, rewarded by executing each task's own verifier by executing each task's own verifier.
Jun-Yao Yang, Yu-Cheng Shi, Zhong-Zhi Li et al.· 9 citations
Reinforcement Learning with Verifiable Rewards (RLVR) has shown remarkable success in improving the mathematical reasoning of large language models. Yet problems beyond the model's current capability, where rollouts uniformly fail and no learning signal is produced, are structurally wasted despite marking the most info...
Yu-Kang Zhu, Zhen-Mao Han· Conference on Empirical Meth...· 0 citations
This work proposes DRACO: Distributing Rubric-based Advantage for Credit Optimization, which generates rubrics dynamically during training to track the policy's evolving capability, scores those rubrics once per completed trajectory, and redistributes that judgment over the steps responsible for annotated rubrics to pr...
Shubham Gandhi, Saurabh Goyal, K. Kate et al.· 5 citations
Gradient-Aligned Reward (GAR), which operates in the policy's own gradient space: truncated backpropagation through the output projection layer extracts a compact gradient vector for each rollout, and cosine similarity with an expert-anchor gradient yields a dense, reasoning-aware reward with less than 9% wall-clock ov...
Le-Qi Zheng, Jin-Bo Su, Fang Niu et al.· 2 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
Microsoft Research Blog· microsoft.comNov 16, 2026
AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduNov 13, 2026