Breaking the Correctness-Efficiency Pareto Frontier in RLVR for Code
francoisfleuret · x · 2026-07-30
A new study tackles combining code correctness and execution efficiency in online Reinforcement Learning (RL), successfully breaking the Pareto frontier between the two objectives.
The research highlights that simply adding execution time to the reward function causes RL to fail due to measurement noise and reward sparsity. To solve this, the authors propose a multi-stage approach: building a calibrated sandbox with large optimization tests, composing speed and correctness rewards effectively, and adapting GRPO to handle sparse, noisy timed-execution settings.
Experiments on DMC-Optim show significant gains, improving strict top-50% pass@1 from 18.0% to 31.3% on Qwen 2.5 7B. CWM 32B sees a 125% relative improvement while preserving pure-correctness scores.
Related event: Meta Research Breaks Pareto Frontier in Code Efficiency via RL(3 posts)→
More from Research
- NeurIPS Reviewer Ghosting: How to Handle Ignored Rebuttals — grumpket · 2026-07-30
- AIPOCH Open-Sources Library of 550+ Medical Research Agent Skills — tom_doerr · 2026-07-30
- ICSE'26 Paper Proposes New Paradigm: In-vivo Fuzzing Without Test Harnesses — moarbugs · 2026-07-30
- LMSYS Debuts Miles: Blackwell-Native 8-bit and 4-bit RL Recipes — BanghuaZ · 2026-07-30
- NeurIPS 2026 Workshop Tackles Continual Learning in Deployed AI Agents — DanielKhashabi · 2026-07-30
- MONTREAL.AI Releases 53-Page Paper: A Framework for Forecasting an Accelerating AI World — Ghost_Pilot_MD · 2026-07-30