AI Paper Atlas

Research map / Benchmarks / GSM8K

GSM8K leaderboard

583 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on GSM8K. Best accuracy: 99.9 (Ours-ICL, NeurIPS 2025).

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.

GSM8K · accuracy

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

#accuracy ↑MethodPaperTopic
199.9Ours-ICL
DeepSeek-V3
Causal Sufficiency and Necessity Improves Chain-of-Thought Reasoning
NeurIPS 2025 · RQ2 ICL table
Reasoning and chain of thought
299.1TG-RAG
Qwen3-32B
TG-RAG: A Retrieval-Augmented Framework for Reasoning Guidance in Specialized Domains
ICML 2026 · Table 1
Retrieval and knowledge
398RaLU
Qwen2.5-72B
Reasoning-as-Logic-Units: Scaling Test-Time Reasoning in Large Language Models Through Logic Unit Alignment
ICML 2025 · Table 1
Reasoning and chain of thought
497.73Adaptive Think
QwQ-32B
Think or Not? Exploring Thinking Efficiency in Large Reasoning Models via an Information-Theoretic Lens
NeurIPS 2025 · Table 1
Reasoning and chain of thought
597.35SGE
GPT-4
Self-Guiding Exploration for Combinatorial Problems
NeurIPS 2024 · Reasoning tasks table
Agents, code and math
697.33FoT (n=4)
QwQ-32B-preview
Forest-of-Thought: Scaling Test-Time Compute for Enhancing LLM Reasoning
ICML 2025 · Main Results
Reasoning and chain of thought
797GPT4-Code + CSV + voting (5 paths)
GPT4-Code
Solving Challenging Math Word Problems Using GPT-4 Code Interpreter with Code-based Self-Verification
ICLR 2024 · GSM8K results table
Agents, code and math
896.7VeriThinker
Qwen-2.5-Math-7B-Instruct
VeriThinker: Learning to Verify Makes Reasoning Model Efficient
NeurIPS 2025 · Results on Short-CoT LLMs tabl
Reasoning and chain of thought
996.51Q-Value steering
Phi-4-reasoning
Modeling Hierarchical Thinking in Large Reasoning Models
ICML 2026 · Table 1
Reasoning and chain of thought
1096.44MESA
Qwen3-30B-A3B
MESA: Improving MoE Safety Alignment via Decentralized Expertise
ICML 2026 · Table 1
Language, safety and interpretability
459 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

GSM8K · pass@1

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

#pass@1 ↑MethodPaperTopic
196.3SwS-32B
Qwen2.5-32B
SwS: Self-aware Weakness-driven Problem Synthesis in Reinforcement Learning for LLM Reasoning
NeurIPS 2025 · Table 1
Reasoning and chain of thought
295.88MixReasoning
Qwen3-8B
MixReasoning: Switching Modes to Think
ICML 2026 · Table 1
Reasoning and chain of thought
395.3RAST + ΔR14B
Qwen-2.5-32B
RAST: Reasoning Activation in LLMs via Small-model Transfer
NeurIPS 2025 · Table 1
Reasoning and chain of thought
495.2rStar-Math (7B SLM+7B PPM)⁶⁴
Qwen2.5-Math-7B
rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking
ICML 2025 · Table: results on challenging
Reasoning and chain of thought
594.39Step-entropy pruning (80% low-entropy)
Qwen3-8B
Making Slow Thinking Faster: Compressing LLM Chain-of-Thought via Step Entropy
ICLR 2026 · Table 1
Reasoning and chain of thought
694.31Self-Harmony
Qwen3-4B-Base
SELF-HARMONY: LEARNING TO HARMONIZE SELF-SUPERVISION AND SELF-PLAY IN TEST-TIME REINFORCEMENT LEARNING
ICLR 2026 · Table 1
Reasoning and chain of thought
794.2Reasoning Scaffolding
Qwen2.5-14B-Ins
Reasoning Scaffolding: Distilling the Flow of Thought from LLMs
ICLR 2026 · Table 1
Reasoning and chain of thought
893.69Dr. MAMR
Qwen2.5-14B-Instruct
Unlocking the Power of Multi-Agent LLM for Reasoning: From Lazy Agents to Deliberation
ICLR 2026 · Table 1
LLM agents
993.6Ours (Manifold Steering)
DeepSeek-R1-Distill-Qwen-14B
Mitigating Overthinking in Large Reasoning Models via Manifold Steering
NeurIPS 2025 · Table 1
Reasoning and chain of thought
1093.11FROST
Phi-4-Reasoning
FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning
ICLR 2026 · Table 1
Reasoning and chain of thought
34 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

GSM8K · score

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

#score ↑MethodPaperTopic
198.2AoT
DeepSeek-V3
Atom of Thoughts for Markov LLM Test-Time Scaling
NeurIPS 2025 · Performance Comparison table
Reasoning and chain of thought
296.01FlowBot
GPT-4.1 mini
FlowBot: Inducing LLM Workflows with Bilevel Optimization and Textual Gradients
ICML 2026 · Table 3
Agents, code and math
395.6Qwen2.5-math-7B-Instruct + Countermath-SFT
Qwen2.5-math-7B-Instruct
One Example Shown, Many Concepts Known! Counterexample-Driven Conceptual Reasoning in Mathematical LLMs
ICML 2025 · Out-of-distribution evaluation
Agents, code and math
491.3PMA
Seed-Dense-70B
Model Merging in Pre-training of Large Language Models
NeurIPS 2025 · text
Language, safety and interpretability
590.07InfiFPO
Phi-4
InfiFPO: Implicit Model Fusion via Preference Optimization in Large Language Models
NeurIPS 2025 · Table Main Results
Alignment and preferences
683.9UniGist
Llama-3.1-8B-Instruct
UniGist: Towards General and Hardware-aligned Sequence-level Long Context Compression
NeurIPS 2025 · Short-context table
Attention and transformers
783.2IF-RLVR Tulu-DPO 8B
Tulu-DPO
Generalizing Verifiable Instruction Following
NeurIPS 2025 · Table: Model performances on d
Agents, code and math
882.51+CCD-DS
Dream-7B-Instruct
Beyond Confidence: Adaptive and Coherent Decoding for Diffusion Language Models
ICML 2026 · Table 1
Fast diffusion sampling
981.6Dream-Instruct + G-Star
Dream-Instruct 7B
Guided Star-Shaped Masked Diffusion
ICML 2026 · Table 1
Fast diffusion sampling
1080.8INPO
LLaMA-3-8B SFT
Iterative Nash Policy Optimization: Aligning LLMs with General Preferences via No-Regret Learning
ICLR 2025 · Table 2
RL from feedback
33 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

GSM8K · AUROC

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

#AUROC ↑MethodPaperTopic
192.7Self-Random (M=5) + Consistency
GPT-3.5
Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs
ICLR 2024 · Table: Comparison of sampling
Language, safety and interpretability
290.37ARS (CCS)
Qwen3-8B
Harnessing Reasoning Trajectories for Hallucination Detection via Answer-agreement Representation Shaping
ICML 2026 · Table 1
Language, safety and interpretability
389.7Input Clarification Ensembling
gpt-3.5-turbo-0613
Decomposing Uncertainty for Large Language Models through Input Clarification Ensembling
ICML 2024 · Table: Uncertainty quantificat
Language, safety and interpretability
485.2Q-Function (SGS)
OpenMath-CodeLlama-7b-Python-hf
Learning Goal-Conditioned Representations for Language Reward Models
NeurIPS 2024 · Table: Train-time goal state a
RL from feedback
580.61TRACED
DeepSeek-R1-Llama-8B
Beyond Scalars: Evaluating and Understanding LLM Reasoning via Geometric Progress and Stability
ICML 2026 · Table 1
Language, safety and interpretability
680.6fDBD (all tokens)
Qwen-3-32B
From Out-of-Distribution Detection to Hallucination Detection: A Geometric View
ICML 2026 · Table 9
Language, safety and interpretability
778.92GIBS
Phi4-reasoning
Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution
ICML 2026 · Table 1
Reasoning and chain of thought
877.28GeoR
R1-Qwen-1.5B
The Geometry of Reasoning: Self-Evaluation via Layerwise Trajectory Evolution
ICML 2026 · Table 1
Reasoning and chain of thought
976ReVISE
Llama-3.2-1B
ReVISE: Learning to Refine at Test-Time via Intrinsic Self-Verification
ICML 2025 · Quantify the Verifying Perform
Reasoning and chain of thought
1070.17CRV
Llama 3.1 8B Instruct
Verifying Chain-of-Thought Reasoning via Its Computational Graph
ICLR 2026 · Table 1
Reasoning and chain of thought
3 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

GSM8K · EM

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

#EM ↑MethodPaperTopic
195.5PDA-GAM (3 deltas)
Qwen3-8B
From “Weak” Signals to Strong Models: Preference Delta Aggregation with LoRA Merging
NeurIPS 2026 · Table 1
Alignment and preferences
291.2SMILE
Qwen3-4B-Instruct
SMILE: Extended Deep Submodular Function-Based Instruction and In-context Learning Demonstration Selection
ICML 2026 · Table 1
Instruction and fine-tuning
391.13CASE
GPT-4o-mini
Sample Efficient Demonstration Selection for In-Context Learning
ICML 2025 · Table 1
Language, safety and interpretability
484.5DPO-D(CP+TP), β=0.7
Qwen2.5-3B-Instruct
On Extending Direct Preference Optimization to Accommodate Ties
NeurIPS 2025 · Table: Preservation Rate (PR%)
Prediction and decision losses
580.74TLM
Llama3-8B-Instruct
Test-Time Learning for Large Language Models
ICML 2025 · Table: Comparison experimental
Instruction and fine-tuning
678.2CRITIC
ChatGPT (gpt-3.5-turbo)
CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing
ICLR 2024 · Table 2
Agents, code and math
777.2BRIDGE
GPT3.5
Exploring Equation as a Better Intermediate Meaning Representation for Numerical Reasoning of Large Language Models
AAAI 2024 · Table 1
Agents, code and math
844.5RoSTE
Llama 3.1 8B
RoSTE: An Efficient Quantization-Aware Supervised Fine-Tuning Approach for Large Language Models
ICML 2025 · Table: Results on Exp.2
Instruction and fine-tuning
939.34Ours (7B transfer)
LLaMA2-13B
On Giant's Shoulders: Effortless Weak to Strong by Dynamic Logits Fusion
NeurIPS 2024 · Table 1
Instruction and fine-tuning
1034.34IRFT T=10
zephyr-7b-sft-full
Getting More Juice Out of the SFT Data: Reward Learning from Human Demonstration Improves SFT for LLM Alignment
NeurIPS 2024 · Table 2
RL from feedback
2 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

GSM8K · speedup

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

#speedup ↑MethodPaperTopic
14.48EAGLE-3 with fused features
LLaMA-Instruct 3.1 8B
EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test
NeurIPS 2025 · Table 2
Reasoning and chain of thought
24.33DAWN
LLaDA-8B-Instruct
DAWN: Dependency-Aware Fast Inference for Diffusion LLMs
NeurIPS 2026 · Table 1
Language, safety and interpretability
33.2SlowFast Sampling
LLaDA 8B
Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principles
ICLR 2026 · Table 1
Language, safety and interpretability
43.13Falcon
Vicuna-13B
Falcon: Faster and Parallel Inference of Large Language Models Through Enhanced Semi-Autoregressive Drafting and Custom-Designed Decoding Tree
AAAI 2025 · Table 1
Reasoning and chain of thought
53CLLM-LLaMA2-7B + Jacobi
LLaMA2-7B
CLLMs: Consistency Large Language Models
ICML 2024 · Table 1
Language, safety and interpretability
62.71EB-Sampler, entropy, γ=0.1
Dream 7B
Accelerated Sampling from Masked Diffusion Models via Entropy Bounded Unmasking
NeurIPS 2025 · Appendix table: Dream 7B GSM8K
Fast diffusion sampling
72.66CTC-drafter
Vicuna-13B
Speculative Decoding with CTC-based Draft Model for LLM Inference Acceleration
NeurIPS 2024 · Table: performance of average
Reasoning and chain of thought
82.11SD+LR
DeepSeek-R1-Distill 32B
Scaling Speculative Decoding with Lookahead Reasoning
NeurIPS 2025 · Performance Across Datasets ta
Reasoning and chain of thought
92.05L_D M + λ L_N-gram (68M)
Llama-68M → Qwen2.5 32B
OmniDraft: A cross-vocabulary, online adaptive drafter for on-device speculative decoding
NeurIPS 2025 · Scalability table
Reasoning and chain of thought
101.63FAFO-Stream
Llama-2-7b-chat
FAFO: Lossy KV Cache Compression for Lossless Inference Acceleration via Draftless Fumble Decoding
ICML 2026 · Table 2
RNNs and normalization
2 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.