AI Paper Atlas

Research map / Benchmarks / NQ

NQ leaderboard

55 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on NQ. Best EM: 72 (CoopRAG (GPT-4o-mini), 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.

NQ · EM

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

#EM ↑MethodPaperTopic
172CoopRAG (GPT-4o-mini)
GPT-4o-mini
Cooperative Retrieval-Augmented Generation for Question Answering: Mutual Information Exchange and Ranking by Contrasting Layers
NeurIPS 2025 · Table: QA performance comparis
Retrieval and knowledge
257.83UniGen-Iter
T5-base
UniGen: A Unified Generative Framework for Retrieval and Question Answering with Large Language Models
AAAI 2024 · Table 2
Visual question answering
357HierSearch
Qwen2.5-7B-Instruct
HierSearch: A Hierarchical Enterprise Deep Search Framework Integrating Local and Web Searches
AAAI 2026 · Table 1
LLM agents
456.1BTR-Atlas large
Atlas large
BTR: Binary Token Representations for Efficient Retrieval Augmented Language Models
ICLR 2024 · Main Results table
Retrieval and knowledge
554.8SkillOrchestra+
Qwen2.5-3B
SkillOrchestra: Learning to Route Agents via Skill Transfer
NeurIPS 2026 · Table 1
LLM agents
654GenRead (FiD-xl) merged with retrieved docs
InstructGPT
Generate rather than Retrieve: Large Language Models are Strong Context Generators
ICLR 2023 · Table 3
Retrieval and knowledge
752.5EMDR² (MSS retriever, MSS reader)
T5-base reader; dual-encoder retriever
End-to-End Training of Multi-Document Reader and Retriever for Open-Domain Question Answering
NeurIPS 2021 · Main results table
Molecules and proteins
852.11Search-R1-GRPO-TRM
Qwen2.5-7B-Instruct
Empowering LLM Tool Invocation with Tool-call Reward Model
ICLR 2026 · Table 1
Reasoning and chain of thought
951.2GRPO + LLDS
Qwen2.5-7B-Base
On Group Relative Policy Optimization Collapse in Agent Search: The Lazy Likelihood-Displacement
ICML 2026 · Table 2
RL from feedback
1050.9Search-R2
Qwen2.5-32B
Search-R2: Enhancing Search-Integrated Reasoning via Actor-Refiner Collaboration
ICML 2026 · Table 3
Reasoning and chain of thought
21 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

NQ · accuracy

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

#accuracy ↑MethodPaperTopic
191vec2vec
GTE → GTR
Harnessing the Universal Geometry of Embeddings
NeurIPS 2025 · Table: in-distribution transla
Recognition and long-tail
282.8WebSeer (Web Search)
Qwen2.5-14B
WebSeer: Training Deeper Search Agents through Reinforcement Learning with Self-Reflection
ICLR 2026 · Table 1
Human-AI and goal-conditioned
379.1MAS2
Qwen3-8B meta-agents
MAS$^2$: Self-Generative, Self-Configuring, Self-Rectifying Multi-Agent Systems
ICLR 2026 · Table 1
LLM agents
472.5KVLink5
Llama-3.1-8B
KVLink: Accelerating Large Language Models via Efficient KV Cache Reuse
NeurIPS 2025 · Main results table
Reasoning and chain of thought
565.9C-3PO
Qwen2-72B
C-3PO: Compact Plug-and-Play Proxy Optimization to Achieve Human-like Retrieval-Augmented Generation
ICML 2025 · Main results table
LLM agents
660.4Tulu3-block-ft
Llama-3.1-Tulu-3-8B-SFT
Block-Attention for Efficient Prefilling
ICLR 2025 · Table 1
Attention and transformers
752.1RAG-DDR (2 rounds)
Llama3-8B
RAG-DDR: Optimizing Retrieval-Augmented Generation Using Differentiable Data Rewards
ICLR 2025 · Table 1
Retrieval and knowledge
846.7AutoRefine-Base
Qwen2.5-3B-Base
Search and Refine During Think: Facilitating Knowledge Refinement for Improved Retrieval-Augmented Reasoning
NeurIPS 2025 · Table 1
Retrieval and knowledge
939.8BTR-BERT base
BERT base
BTR: Binary Token Representations for Efficient Retrieval Augmented Language Models
ICLR 2024 · BERT results table
Retrieval and knowledge
1037.4CalibRAG
Llama-2-7B
Reliable Decision‑Making via Calibration‑Oriented Retrieval‑Augmented Generation
NeurIPS 2025 · Table: Comparison of RAG varia
Retrieval and knowledge
5 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

NQ · F1

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

#F1 ↑MethodPaperTopic
184.21Ours
LLaMA3.1-8B
Fine-Grained Privacy Extraction from Retrieval-Augmented Generation Systems by Exploiting Knowledge Asymmetry
ICLR 2026 · Table 2
Retrieval and knowledge
267.4PANINI
GPT-4o-mini answer model
Panini: Continual Learning in Token Space via Structured Memory
ICML 2026 · Table 2
Retrieval and knowledge
367.24UniGen-Iter
T5-base
UniGen: A Unified Generative Framework for Retrieval and Question Answering with Large Language Models
AAAI 2024 · Table 2
Visual question answering
463.58Psi-RAG (keyword abstract)
Llama-3.3-70B
Hierarchical Abstract Tree for Cross-Document Retrieval Augmented Generation
ICML 2026 · Table 2
Human and embodied modeling
559.45JADE
Qwen2.5-7B-Instruct
JADE: Bridging the Strategic-Operational Gap in Dynamic Agentic RAG
ICML 2026 · Table 2
LLM agents
656.37ExSearch-Qwen2.5-7B
Qwen2.5-7B
Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers
NeurIPS 2025 · Table 1
Retrieval and knowledge
752.5DECOR
Qwen3-8B-Instruct
DECOR: Learning to Decompose and Collaborate in Deep Search via Multi-Agent Reinforcement Learning
ICML 2026 · Table 1
Cooperative multi-agent RL
850Our Method (TTS)
gemini-2.5-flash-lite
Incentive-Aligned LLM Summaries
ICLR 2026 · Table 1
Language, safety and interpretability
949.87Graph-R1
Qwen2.5-7B-Instruct
Graph-R1: Towards Agentic GraphRAG Framework via End-to-end Reinforcement Learning
ICML 2026 · Table 1
Retrieval and knowledge
1049.5CausalRAG2
gpt-5-nano
CausalRAG2: Hierarchical Causal Knowledge Graph Design for RAG
ICML 2026 · Table 4
Knowledge and temporal graphs
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

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