Tsinghua and Tencent Unveil TreeOPSD Boosting Deep Research Accuracy by 4.97 Points

1 reports archived1 independent sourcesupdated 11/12/2026, 21:28:00

Synthesis & Latest Updates

1 Sources Cross-Validated

Researchers from Tsinghua University and Tencent propose TreeOPSD, which improves credit assignment in long-horizon agent reinforcement learning via tree-based trajectory sampling and turn-level credit estimation. On Qwen3-30B-A3B-Thinking-2507, it achieves 21.20% average Exact Match across four Deep Research benchmarks, a 4.97-point gain over GRPO's 16.23%.

LATEST/Researchers from Tsinghua University and Tencent propose TreeOPSD, which improves credit assignment in long-horizon agent reinforcement learning via tree-based trajectory sampling and turn-level credit estimation. On Qwen3-30B-A3B-Thinking-2507, it achieves 21.20% average Exact Match across four Deep Research benchmarks, a 4.97-point gain over GRPO's 16.23%.

Hourly heat curve

20 fully observed hours

Now

7.7

Peak

11.922:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

量子位 (微信公众号)T2Nov 12, 21:28·72 pts
Tsinghua and Tencent Unveil TreeOPSD Boosting Deep Research Accuracy by 4.97 Points
- TreeOPSD uses tree-based trajectory sampling with shared prefixes to back-propagate terminal rewards to intermediate nodes and estimate turn-level credit.
- On Qwen3-30B-A3B-Thinking-2507, it achieves 21.20% average Exact Match across four Deep Research benchmarks, a 4.97-point improvement over GRPO.
- The method also addresses policy staleness by letting the model decide update magnitude based on current state.
View Related Reports & Analysis Original Source