AI Paper Atlas

Research map / Benchmarks / TruthfulQA

TruthfulQA leaderboard

64 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on TruthfulQA. Best accuracy: 91.4 (ABCA, AAAI 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.

TruthfulQA · accuracy

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

#accuracy ↑MethodPaperTopic
191.4ABCA
GPT-4.1
Hallucinate Less by Thinking More: Aspect-Based Causal Abstention for Large Language Models
AAAI 2026 · Table 1
Language, safety and interpretability
290.45CAMP
Qwen2.5-7B-Instruct
Conflict-Aware Adaptive Alignment for LLM Hallucination Mitigation
ICML 2026 · Table 4
Alignment and preferences
386.8BlueAgentBreeder argmax combined (GPQA run)AgentBreeder: Mitigating the AI Safety Risks of Multi-Agent Scaffolds via Self-Improvement
NeurIPS 2025 · Table 2
LLM agents
483.7DataGen-generated dataset
ChatGPT
DataGen: Unified Synthetic Dataset Generation via Large Language Models
ICLR 2025 · Main results on generated and
Agents, code and math
583.5DEAR
GPT-4o-mini
Group Perspective Matters: Regulating Debate Relationships Can Mitigate Blind Conformity in Multi-Agent Debate
NeurIPS 2026 · Table 1
LLM agents
674.3Debate
Gemini-2.0-Flash
Multi-Agent Debate for LLM Judges with Adaptive Stability Detection
NeurIPS 2025 · Comparative Results table
LLM agents
773.4Adaptive Distraction
Claude-3.5-Sonnet
Adaptive Distraction: Probing LLM Contextual Robustness with Automated Tree Search
NeurIPS 2025 · Main results table
Retrieval and knowledge
871.27AlphaPO
Mistral-7B-Instruct
AlphaPO: Reward Shape Matters for LLM Alignment
ICML 2025 · Table 2
Alignment and preferences
969Ours (6× GPT-3.5)
GPT-3.5
Multi-LLM Debate: Framework, Principals, and Interventions
NeurIPS 2024 · Table 1
LLM agents
1068.71IGLB
True/False softmax score
Multicalibration for Confidence Scoring in LLMs
ICML 2024 · Table 1
Language, safety and interpretability
31 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

TruthfulQA · AUROC

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

#AUROC ↑MethodPaperTopic
195.6SE_SDLG
OPT-13b
Improving Uncertainty Estimation through Semantically Diverse Language Generation
ICLR 2025 · Table 1
Language, safety and interpretability
294Diversity-steered sampling
LLaDA 8B-Instruct
Efficient semantic uncertainty quantification in language models via diversity-steered sampling
NeurIPS 2025 · Table: AUROC of SE
Language, safety and interpretability
388.5HARP
LLaMA-3.1-8B
HARP: Hallucination Detection via Reasoning Subspace Projection
ICLR 2026 · Table 1
Language, safety and interpretability
487.3TSV
Qwen-2.5-7B
Steer LLM Latents for Hallucination Detection
ICML 2025 · Main results table
Language, safety and interpretability
586.64ARS (CCS)
Qwen3-8B
Harnessing Reasoning Trajectories for Hallucination Detection via Answer-agreement Representation Shaping
ICML 2026 · Table 1
Language, safety and interpretability
683.68PALE
Qwen-2.5-7B
Bolster Hallucination Detection via Prompt-Guided Data Augmentation
AAAI 2026 · Table 1
Language, safety and interpretability
782.41HaloScope
OPT-13B
HaloScope: Harnessing Unlabeled LLM Generations for Hallucination Detection
NeurIPS 2024 · Larger LLMs table
Language, safety and interpretability
879.4Probe(EOS)
Qwen2.5-7B
Can LLMs Detect Their Confabulations? Estimating Reliability in Uncertainty-Aware Language Models
AAAI 2026 · Table 1
Language, safety and interpretability
972.72HybridGrad
Mistral-Nemo-Instruct-12B
Gradients with Respect to Semantics Preserving Embeddings Tell the Uncertainty of Large Language Model
ICML 2026 · Table 1
Language, safety and interpretability
1072.61SAGE
Qwen3
Detecting Errors in AI-Generated Annotations: When and Why Semantic Neighbors Help
ICML 2026 · Table 1
Language, safety and interpretability
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

TruthfulQA · score

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

#score ↑MethodPaperTopic
164.9INPO
LLaMA-3-8B SFT
Iterative Nash Policy Optimization: Aligning LLMs with General Preferences via No-Regret Learning
ICLR 2025 · Table 2
RL from feedback
262.2UltraFeedback Best/Worst DPO
Llama-3.1-Tulu-3-8B-SFT
What Matters in Data for DPO?
NeurIPS 2025 · Table 1
Alignment and preferences
352.26Patched Mistral-7B-Instruct-v0.1
Mistral-7B-Instruct-v0.1
Interpreting the Repeated Token Phenomenon in Large Language Models
ICML 2025 · Table: The effect of patching
Language, safety and interpretability
450.55WizardLM-13b
LLaMA 13B
WizardLM: Empowering Large Pre-Trained Language Models to Follow Complex Instructions
ICLR 2024 · Table 1
Instruction and fine-tuning
547Qwen2-7B-Instruct + WSPO
Qwen2-7B-Instruct
Weak-to-Strong Preference Optimization: Stealing Reward from Weak Aligned Model
ICLR 2025 · Reasoning benchmark table
Alignment and preferences
646.95MRACL
Qwen2.5-7B-Base
MRACL: Multi-Reward Space Guided Adaptive Curriculum Reinforcement Learning for LLMs
AAAI 2026 · Table 2
RL from feedback
744.95BitDelta
Llama 2-7B
BitDelta: Your Fine-Tune May Only Be Worth One Bit
NeurIPS 2024 · Table 1
Instruction and fine-tuning
843.81PRO-P
Mistral-7B-sft
Proximalized Preference Optimization for Diverse Feedback Types: A Decomposed Perspective on DPO
NeurIPS 2025 · Table: Results of aligning Mis
Alignment and preferences
943.1MMaDA-8B
LLaDA-8B-Instruct
MMaDA: Multimodal Large Diffusion Language Models
NeurIPS 2025 · Table: Evaluation on LLM Bench
Multimodal LLMs
1038.2YaRN 7B, s=16
Llama 2 7B
YaRN: Efficient Context Window Extension of Large Language Models
ICLR 2024 · Table: Standardized Benchmarks
Instruction and fine-tuning

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