DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards
About
Reinforcement learning from verifiable rewards (RLVR) has emerged as a central technique for improving the reasoning capabilities of large language models. Despite its effectiveness, how response-level rewards translate into token-level probability changes remains poorly understood. We introduce a discriminator view of RLVR updates, showing that the policy-gradient update direction implicitly acts as a linear discriminator over token-gradient vectors and thereby determines which token probabilities are increased or decreased during learning. Under standard sequence-level RLVR, this discriminator is constructed from positive- and negative-side centroids formed by advantage-weighted averaging of token-gradient vectors. However, such centroid construction can be dominated by shared high-frequency patterns, such as formatting tokens, diluting sparse yet discriminative directions that better distinguish high-reward responses from low-reward ones. To address this limitation, we propose $\textbf{DelTA}$, a discriminative token credit assignment method that estimates token coefficients to amplify side-specific token-gradient directions and downweight shared or weakly discriminative ones. These coefficients reweight a self-normalized RLVR surrogate, making the effective side-wise centroids more contrastive and thereby reshaping the RLVR update direction. On seven mathematical benchmarks, DelTA outperforms the strongest same-scale baselines by 3.26 and 2.62 average points on Qwen3-8B-Base and Qwen3-14B-Base, respectively. Additional results on code generation, a different backbone, and out-of-domain evaluations further demonstrate the generalization ability of DelTA.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Mathematical Reasoning | AIME 2024 | Accuracy56.87 | 220 | |
| Mathematical Reasoning | AIME 2025 | Accuracy37.92 | 214 | |
| Mathematical Reasoning | AIME 2026 | AIME 2026 Accuracy45.21 | 55 | |
| Mathematical Reasoning | HMMT Feb 2025 | Accuracy26.04 | 45 | |
| Mathematical Reasoning | HMMT Feb 2026 | Accuracy26.89 | 40 | |
| Mathematical Reasoning | HMMT Nov 2025 | -- | 32 | |
| Code Generation | LiveCodeBench (LCB) | Accuracy35.6 | 24 | |
| Mathematical Reasoning | BRUMO 2025 | Accuracy54.79 | 10 | |
| Reasoning | GPQA Diamond | Avg@556.67 | 4 | |
| Reasoning | MMLU-Pro 600 sampled questions | Average @572.27 | 4 |