OpenDeepThink: Parallel Reasoning via Bradley-Terry Aggregation

• URL: http://arxiv.org/abs/2605.15177v2
• Date: Sun, 17 May 2026 19:31:24 GMT
• Title: OpenDeepThink: Parallel Reasoning via Bradley-Terry Aggregation
• Abstract summary: We introduce OpenDeepThink, a population-based test-time compute framework that selects via pairwise Bradley-Terry comparison.<n>OpenDeepThink raises Gemini 3.1 Pro's effective Codeforces Elo by +405 points in eight sequential LLM-call rounds.<n>We release CF-73, a curated set of 73 expert-rated Codeforces problems with International Grandmaster annotation and 99% local-evaluation agreement against the official verdict.
• Score: 53.88666485159289
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Test-time compute scaling is a primary axis for improving LLM reasoning. Existing methods primarily scale depth by extending a single reasoning trace. 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, we introduce OpenDeepThink, a population-based test-time compute framework that selects via pairwise Bradley-Terry comparison. Each generation, the LLM judges random pairs of candidates and aggregates votes via Bradley-Terry into a global ranking; top-ranked candidates are preserved and the top three quarters are mutated using the natural-language critiques produced during comparison; the bottom quarter is discarded. OpenDeepThink raises Gemini 3.1 Pro's effective Codeforces Elo by +405 points in eight sequential LLM-call rounds (~27 minutes wall-clock). The pipeline transfers across weaker and stronger models without retuning, and on the multi-domain HLE benchmark, gains appear concentrated in objectively verifiable domains and reverse in subjective ones. We release CF-73, a curated set of 73 expert-rated Codeforces problems with International Grandmaster annotation and 99% local-evaluation agreement against the official verdict.

Related papers

• Inferring Code Correctness from Specification [0.0]
  Large language models (LLMs) have become integral to modern software development, enabling automated code generation at scale.<n>We propose TRAILS (Targeted Reasoning Agreement via Inputs and Specifications), an approach that grounds LLM reasoning with concrete (input, output) pairs.<n>We evaluate TRAILS on two datasets, LiveCodeBench and CoCoClaNeL, across three LLMs (Qwen3Coder-30B, Devstral-Small-24B, and Olmo3.1-Instruct), comparing against HoarePrompt and a Zero-Shot Chain-of-Thought baseline.
  arXiv  Detail & Related papers  (2026-05-28T12:04:51Z)
• Semantic Voting: Execution-Grounded Consensus for LLM Code Generation [11.731523303184472]
  Best execution-based selector exceeds output-pattern majority voting by 19-52 percentage points on every configuration.<n>Deeper thinking improves majority voting by 12 pp but execution-based methods stay flat or degrade as candidate diversity falls.<n>These results frame inference-time code selection as a signal-quality problem rather than an aggregation-rule problem.
  arXiv  Detail & Related papers  (2026-05-09T04:33:39Z)
• Duel-Evolve: Reward-Free Test-Time Scaling via LLM Self-Preferences [7.715856473694668]
  We introduce Duel-Evolve, an evolutionary algorithm that replaces external scalar rewards with pairwise preferences elicited from the same LLM used to generate candidates.<n>Duel-Evolve aggregates noisy candidate comparisons via a Bayesian Bradley-Terry model, yielding uncertainty-aware estimates of candidate quality.<n>We evaluate Duel-Evolve on MathBench, where it achieves 20 percentage points higher accuracy over existing methods and baselines, and on LiveCodeBench, where it improves over comparable iterative methods by over 12 percentage points.
  arXiv  Detail & Related papers  (2026-02-25T05:16:11Z)
• Beyond Majority Voting: LLM Aggregation by Leveraging Higher-Order Information [57.397381631496906]
  We develop two new aggregation algorithms called Optimal Weight (OW) and Inverse Surprising Popularity (ISP)<n>Our theoretical analysis shows these methods provably mitigate inherent limitations of majority voting under mild assumptions.<n>We empirically validate our algorithms on synthetic datasets, popular LLM fine-tuning benchmarks such as UltraFeedback and MMLU, and a real-world healthcare setting ARMMAN.
  arXiv  Detail & Related papers  (2025-10-01T22:21:50Z)
• Adaptive Repetition for Mitigating Position Bias in LLM-Based Ranking [12.134014941104613]
  Order of candidate items can influence the model's final decision.<n>This sensitivity to item positioning in a LLM's prompt is known as position bias.<n>We introduce a dynamic early-stopping method that adaptively determines the number of repetitions required for each instance.
  arXiv  Detail & Related papers  (2025-07-23T09:54:44Z)
• LiveCodeBench Pro: How Do Olympiad Medalists Judge LLMs in Competitive Programming? [88.29001498765629]
  Large language models (LLMs) now outperform elite humans in competitive programming.<n>We revisit this claim, examining how LLMs differ from human experts and where limitations still remain.<n>We introduce LiveCodeBench Pro, a benchmark composed of problems from Codeforces, ICPC, and IOI.<n>A team of Olympiad medalists annotates every problem for algorithmic categories and conducts a line-by-line analysis of failed model-generated submissions.
  arXiv  Detail & Related papers  (2025-06-13T16:29:09Z)
• GRP: Goal-Reversed Prompting for Zero-Shot Evaluation with LLMs [14.906150451947443]
  Using Large Language Models (LLMs) to evaluate and compare two answers typically involves having LLM-based judges select the better answer.<n>We propose a Goal-Reversed Prompting (GRP) approach for pairwise evaluation that shifts the original task from selecting the better answer to choosing the worse one.
  arXiv  Detail & Related papers  (2025-03-08T09:44:24Z)
• Language Models are Hidden Reasoners: Unlocking Latent Reasoning Capabilities via Self-Rewarding [74.31981011985681]
  Large language models (LLMs) have shown impressive capabilities, but still struggle with complex reasoning tasks requiring multiple steps. We introduce LaTent Reasoning Optimization (LaTRO), a principled framework that formulates reasoning as sampling from a latent distribution. We validate LaTRO through experiments on GSM8K and ARC-Challenge datasets using multiple model architectures.
  arXiv  Detail & Related papers  (2024-11-06T22:02:30Z)
• Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking [56.275521022148794]
  Post-training methods claim superior alignment by virtue of better correspondence with human pairwise preferences.<n>We attempt to answer the question -- do LLM-judge preferences translate to progress on other, more concrete metrics for alignment, and if not, why not?<n>We find that (1) LLM-judge preferences do not correlate with concrete measures of safety, world knowledge, and instruction following; (2) LLM-judges have powerful implicit biases, prioritizing style over factuality and safety; and (3) the supervised fine-tuning stage of post-training, and not the PO stage, has the greatest impact on alignment.
  arXiv  Detail & Related papers  (2024-09-23T17:58:07Z)
• Found in the Middle: Permutation Self-Consistency Improves Listwise Ranking in Large Language Models [63.714662435555674]
  Large language models (LLMs) exhibit positional bias in how they use context. We propose permutation self-consistency, a form of self-consistency over ranking list outputs of black-box LLMs. Our approach improves scores from conventional inference by up to 7-18% for GPT-3.5 and 8-16% for LLaMA v2 (70B)
  arXiv  Detail & Related papers  (2023-10-11T17:59:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.