REVIEW 32 cited by
ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning
read the original abstract
We investigate the logical reasoning capabilities of large language models (LLMs) and their scalability in complex non-monotonic reasoning. To this end, we introduce ZebraLogic, a comprehensive evaluation framework for assessing LLM reasoning performance on logic grid puzzles derived from constraint satisfaction problems (CSPs). ZebraLogic enables the generation of puzzles with controllable and quantifiable complexity, facilitating a systematic study of the scaling limits of models such as Llama, o1 models, and DeepSeek-R1. By encompassing a broad range of search space complexities and diverse logical constraints, ZebraLogic provides a structured environment to evaluate reasoning under increasing difficulty. Our results reveal a significant decline in accuracy as problem complexity grows -- a phenomenon we term the curse of complexity. This limitation persists even with larger models and increased inference-time computation, suggesting inherent constraints in current LLM reasoning capabilities. Additionally, we explore strategies to enhance logical reasoning, including Best-of-N sampling, backtracking mechanisms, and self-verification prompts. Our findings offer critical insights into the scalability of LLM reasoning, highlight fundamental limitations, and outline potential directions for improvement.
Forward citations
Cited by 32 Pith papers
-
MathConstraint: Automated Generation of Verified Combinatorial Reasoning Instances for LLMs
MathConstraint generates scalable, automatically verifiable combinatorial problems where LLMs achieve 18.5-66.9% accuracy without tools but roughly double that with solver access.
-
Transferability for General Reasoning: An Automated Curriculum for Multi-Domain RLVR
TAC is a new online curriculum for multi-domain RLVR that estimates cross-domain transferability from gradient geometry to achieve up to 2.8 point macro-averaged accuracy gains over learnability-only baselines on six-...
-
Transferability for General Reasoning: An Automated Curriculum for Multi-Domain RLVR
TAC is a bandit curriculum for multi-domain RLVR that prioritizes domains whose gradient updates align with and benefit other domains, yielding up to 2.8-point macro accuracy gains over learnability-only baselines on ...
-
DICE: Entropy-Regularized Equilibrium Selection for Stable Multi-Agent LLM Coordination
DICE formalizes multi-agent LLM coordination as discounted incomplete-information Markov games and introduces Heterogeneous Quantal Response Equilibrium (HQRE) to achieve unique stable equilibria with bounded regret, ...
-
CORE: Contrastive Reflection Enables Rapid Improvements in Reasoning
CORE distills contrasts between successful and unsuccessful reasoning traces into compact natural-language insights that enable faster model self-improvement on reasoning tasks with fewer rollouts than parametric or o...
-
LLMEval-Logic: A Solver-Verified Chinese Benchmark for Logical Reasoning of LLMs with Adversarial Hardening
LLMEval-Logic is a solver-verified Chinese logical reasoning benchmark with 246 base and 190 hard items that shows frontier LLMs reach only 37.5% hard-item accuracy and 60.16% joint formalization score.
-
SUPERNOVA: Eliciting General Reasoning in LLMs with Reinforcement Learning on Natural Instructions
SUPERNOVA adapts instruction-tuning data for RLVR and achieves up to 52.8% relative gains on general reasoning benchmarks like BBEH through targeted task selection and mixing.
-
DICE: Entropy-Regularized Equilibrium Selection for Stable Multi-Agent LLM Coordination
HQRE entropy regularization makes multi-agent LLM coordination well-posed, yielding unique equilibria, linear mirror convergence, bounded Bayesian regret, and DICE gains of 4.3–8.5 pp on reasoning/planning tasks.
-
TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination
Stale-occupancy sequential fine-tuning of multi-agent LLMs incurs an O(n²) certificate penalty; TeamTR resamples under intermediate occupancy and enforces token-level trust regions to restore O(n) scaling and stable gains.
-
CA-BED: Conversation-Aware Bayesian Experimental Design
CA-BED uses Bayesian experimental design and simulated conversation trees with LLM likelihoods to optimize multi-turn question selection, reporting 21.8% higher success rates than direct prompting on entity-deduction ...
-
Forge: Quality-Aware Reinforcement Learning for NP-Hard Optimization in LLMs
OPT-BENCH trains LLMs on NP-hard optimization via quality-aware RLVR, achieving 93.1% success rate and 46.6% quality ratio on Qwen2.5-7B while outperforming GPT-4o and transferring gains to other domains.
-
TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination
TeamTR is a trust-region framework for multi-agent LLM fine-tuning that resamples trajectories after each update to convert quadratic compounding occupancy shift into linear scaling and yields per-update improvement l...
-
Train Separately, Merge Together: Modular Post-Training with Mixture-of-Experts
BAR trains independent domain experts via separate mid-training, SFT, and RL pipelines then composes them with a MoE router to match monolithic retraining performance at lower cost and without catastrophic forgetting.
-
GroupDPO: Memory efficient Group-wise Direct Preference Optimization
GroupDPO decouples group-wise preference optimization during backpropagation to cut peak memory while keeping the same gradients, allowing larger groups and consistent gains over single-pair DPO plus an NLL term on positives.
-
The Format Tax
Structured-output instructions alone impose a large accuracy tax on open-weight LLMs; decoupling freeform reasoning from formatting recovers most of it, while recent closed models largely avoid the tax.
-
Qwen3-Omni Technical Report
Qwen3-Omni is a unified multimodal model that achieves open-source SOTA on 32 of 36 audio and audio-visual benchmarks and overall SOTA on 22 without degrading performance on text, image, or video relative to single-mo...
-
Investigating Advanced Reasoning of Large Language Models via Black-Box Environment Interaction
Introduces the Oracle benchmark of 96 black-box environments across 6 task types to measure integrated reasoning in LLMs through interactive function discovery, with o3 leading but all models showing planning weakness...
-
InternBootcamp Technical Report: Boosting LLM Reasoning with Verifiable Task Scaling
InternBootcamp supplies 1000+ verifiable, auto-generated task environments across domains that enable task scaling to improve LLM reasoning, producing a 32B model with state-of-the-art results on the new Bootcamp-EVAL...
-
MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention
MiniMax-M1 is a 456B parameter hybrid-attention MoE model trained with CISPO RL that achieves performance comparable or superior to DeepSeek-R1 and Qwen3-235B on reasoning and software engineering tasks while training...
-
AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning
AReaL decouples generation and training in LLM reinforcement learning to achieve up to 2.77x speedup with matched or better performance on math and code benchmarks.
-
SUPERNOVA: Eliciting General Reasoning in LLMs with Reinforcement Learning on Natural Instructions
Curating 25K natural-instruction examples for RLVR yields a 64.4pp relative gain on BBEH for Qwen3-0.6B and generalizes across scales and model families.
-
DOPD: Dual On-policy Distillation
DOPD is an advantage-aware dual distillation method that dynamically assigns token supervision from either privileged teacher or student to transfer capability while mitigating non-replicable information asymmetry in ...
-
Continual LLM Upcycling: A Predictor-Gated Bank-Wise Sparsity Training Recipe for Dense-to-Sparse LLMs
Continual training recipe upcycles dense Qwen2.5-8B LLM to 4x channel-sparse model via predictor-gated bank-wise sparsity in SwiGLU FFN with a single-layer repair for long-context failure on RULER-CWE.
-
Benchmarking EngGPT2-16B-A3B against Comparable Italian and International Open-source LLMs
EngGPT2MoE-16B-A3B matches or exceeds other Italian open-source LLMs on most international benchmarks while remaining competitive on ITALIC, though it trails some top international models.
-
HyperLens: Quantifying Cognitive Effort in LLMs with Fine-grained Confidence Trajectory
HyperLens reveals that deeper transformer layers magnify small confidence changes into fine-grained trajectories, allowing quantification of cognitive effort where complex tasks demand more and standard SFT can reduce it.
-
Where Reasoning Breaks: Logic-Aware Path Selection by Controlling Logical Connectives in LLMs Reasoning Chains
Targeting logical connectives with gradient steering, localized branching, and transition optimization improves LLM reasoning chain stability and efficiency.
-
Kimi K2: Open Agentic Intelligence
Kimi K2 is a 1-trillion-parameter MoE model that leads open-source non-thinking models on agentic benchmarks including 65.8 on SWE-Bench Verified and 66.1 on Tau2-Bench.
-
Qwen3 Technical Report
Pith review generated a malformed one-line summary.
-
NebulaExp-8B: An Empirical Post-Training Pipeline via Full-Scale Ablation Research
NebulaExp reports an empirical post-training pipeline on Qwen3-8B that raises instruct scores from 55.01 to 61.85 and reasoning scores from 73.88 to 75.17 via curated data, SFT, GRPO RL, and OPD/MOPD distillation.
-
The Periodic Table of LLM Reasoning: A Structured Survey of Reasoning Paradigms, Methods, and Failure Modes
A literature survey that introduces a taxonomy for LLM reasoning paradigms, analyzes methodological trends, and synthesizes failure modes from over 300 papers.
-
Soft Adaptive Policy Optimization
SAPO introduces smooth adaptive gating to replace hard clipping in token- and sequence-level policy optimization for more stable LLM reinforcement learning.
-
Benchmarking EngGPT2-16B-A3B against Comparable Italian and International Open-source LLMs
EngGPT2MoE-16B-A3B matches or beats other Italian models on most international benchmarks but trails top international models such as GPT-5 nano and Qwen3-8B.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.