OpenDeepThink: Parallel Test-Time Reasoning via Bradley-Terry Aggregation

Forum topic · 小凯 · 2026-05-17

Summary

OpenDeepThink is a population-based test-time compute framework for improving LLM reasoning, introduced by researchers including Shang Zhou and Jingbo Shang (arXiv:2605.15177). Instead of only deepening a single reasoning chain, it expands breadth by sampling multiple candidate solutions in parallel and addresses the selection bottleneck of noisy pointwise LLM judges through pairwise Bradley-Terry comparisons. Each generation, an LLM judge evaluates random candidate pairs; votes are aggregated into a global Bradley-Terry ranking. The top-ranked candidates are retained, the top three-quarters are mutated using natural-language critiques produced during comparisons, and the bottom quarter is discarded. Across eight sequential LLM call rounds (~27 minutes wall-clock), OpenDeepThink raises Gemini 3.1 Pro's effective Codeforces Elo by +405. The pipeline transfers across weaker and stronger models without retuning. On the multi-domain HLE benchmark, gains concentrate in objectively verifiable domains and reverse in subjective ones. The authors also release CF-73, a set of 73 expert-rated Codeforces problems with grandmaster annotations and 99% agreement with official verdicts in local evaluation.

Overview

Field: Machine Learning Authors: Shang Zhou, Wenhao Chai, Kaiyuan Liu, Huanzhi Mao, Qiuyang Mang, Jingbo Shang Published: 2026-05-14 arXiv: 2605.15177

Abstract (translated)

Test-time compute scaling is a major axis for improving LLM reasoning. Existing methods mainly increase depth by extending a single reasoning trajectory. Scaling breadth by sampling multiple candidates in parallel is straightforward, but introduces a selection bottleneck: choosing the best candidate without a ground-truth verifier, since pointwise LLM judging is noisy and biased.

To address this, the authors introduce OpenDeepThink, a population-based test-time compute framework that performs selection via pairwise Bradley-Terry comparisons. Each generation:

1. An LLM judge evaluates random pairs of candidates. 2. Votes are aggregated via Bradley-Terry aggregation into a global ranking. 3. Top-ranked candidates are retained. 4. The top three-quarters are mutated using natural-language critiques generated during the pairwise comparisons. 5. The bottom quarter is discarded.

Key Results

Across eight sequential rounds of LLM calls (~27 minutes wall-clock time), OpenDeepThink improves Gemini 3.1 Pro's effective Codeforces Elo by +405.
The pipeline transfers across weaker and stronger models without retuning.
On the multi-domain HLE benchmark, gains concentrate in objectively verifiable domains and reverse in subjective ones.

Released Benchmark: CF-73

The team releases CF-73: 73 expert-rated Codeforces problems with grandmaster annotations, achieving 99% agreement with official verdicts in local evaluation.

---

*Auto-collected on 2026-05-17.*