AI Paper Atlas

Research map / Benchmarks / MATH500

MATH500 leaderboard

269 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on MATH500. Best accuracy: 99.3 (TeamTR Homogeneous (3x8B), ICML 2026).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

MATH500 · accuracy

186 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
199.3TeamTR Homogeneous (3x8B)
Qwen3-8B
TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination
ICML 2026 · Table 1
LLM agents
298.4Hermes@1
DeepSeek-V3.2
HERMES: Towards Efficient and Verifiable Mathematical Reasoning in LLMs
ICML 2026 · Table 1
Reasoning and chain of thought
397.4ThoughtFold
Qwen3-8B
ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning
ICML 2026 · Table 1
Reasoning and chain of thought
497.4Sonata
Qwen3-8B
Adaptive Thinking: Large Language Models Know When to Think in Latent Space
ICLR 2026 · Table 1
Reasoning and chain of thought
596.8SpecExit
Qwen3-4B-Thinking-2507
SpecExit: Accelerating Large Reasoning Model via Speculative Exit
ICML 2026 · Table 1
Reasoning and chain of thought
696.8QwQ-32B w/ TrimR
QwQ-32B
TrimR: Verifier-based Training-Free Thinking Trimming for Efficient Test-Time Scaling
ICLR 2026 · Table 1
Reasoning and chain of thought
796.8Ours (latent QP projection)
Qwen3-14B
Efficient Reasoning via Constrained Optimization in Latent Space
NeurIPS 2026 · Table 1
Reasoning and chain of thought
896.6Prompt-Hint-SFT-32B
DeepSeek-R1-32B
Teaching Language Models to Reason with Tools
NeurIPS 2025 · Table 1
Reasoning and chain of thought
996.25Greedy LinUCB
6-LLM pool
Online Multi-LLM Selection via Contextual Bandits Under Unstructured Context Evolution
AAAI 2026 · Table 1
Contextual bandits and Thompson sampling
1096.2Ours-ICL
DeepSeek-V3
Causal Sufficiency and Necessity Improves Chain-of-Thought Reasoning
NeurIPS 2025 · RQ2 ICL table
Reasoning and chain of thought
176 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

MATH500 · pass@1

85 papers, each with its best reported number for its own method; higher is better.

#pass@1 ↑MethodPaperTopic
198.3CRD-30B-A3B
Qwen3-30B-A3B
Collaborative Reasoning Distillation via Cross-Feedback and Coherent Curation
NeurIPS 2026 · Table 1
Reasoning and chain of thought
297.4Adaptive control
QwQ-32B
Controlling Thinking Speed in Reasoning Models
NeurIPS 2025 · Adaptive control results table
Reasoning and chain of thought
397.2Soft (Gumbel)
QwQ-32B
LLMs are Single-threaded Reasoners: Demystifying the Working Mechanism of Soft Thinking
ICLR 2026 · Table 2
Reasoning and chain of thought
496.4PADD
Qwen3-30B-A3B student
PADD: Path-Aligned Decompression Distillation for Non-Router Teacher to Guide MoE Student Learning
ICML 2026 · Table 1
Reasoning and chain of thought
596.2DyCon (Ours)
Qwen3-4B-Thinking-2507
DyCon: Dynamic Reasoning Control via Evolving Difficulty Modeling
ICML 2026 · Table 1
Reasoning and chain of thought
696.2AdvChain (Ours)
Qwen3-4B
AdvChain: Adversarial Chain-of-Thought Tuning for Robust Safety Alignment of Large Reasoning Models
ICLR 2026 · Table 4
Reasoning and chain of thought
795.2ReBalance
QwQ-32B
Efficient Reasoning with Balanced Thinking
ICLR 2026 · Table 1
Reasoning and chain of thought
895+SAGE-GRPO
Qwen3-8B
Does Your Reasoning Model Implicitly Know When to Stop Thinking?
ICML 2026 · Table 1
Reasoning and chain of thought
994.8DS-R1-Distill-Qwen-7B + GAR
DeepSeek-R1-Distill-Qwen-7B
Generative Adversarial Reasoner: Enhancing LLM Reasoning with Adversarial Reinforcement Learning
ICLR 2026 · Table 1
Reasoning and chain of thought
1093.8CRAFT
Qwen3-4B-Thinking
Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations
ICML 2026 · Table 2
Reasoning and chain of thought
75 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.