AI Paper Atlas

Research map / Benchmarks / IFEval

IFEval leaderboard

32 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on IFEval. Best score: 85.4 (RemeDi (++ Remask RL), ICLR 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.

IFEval · score

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

#score ↑MethodPaperTopic
185.4RemeDi (++ Remask RL)
LLaDA-8B-Instruct
Don't Settle Too Early: Self-Reflective Remasking for Diffusion Language Models
ICLR 2026 · Table 2
Fast diffusion sampling
285.4Llama-3.1-8B-Instruct + Paprika
Llama-3.1-8B-Instruct
Training a Generally Curious Agent
ICML 2025 · Table: Evaluation of Paprika o
Human-AI and goal-conditioned
382.25InfiFPO
Phi-4
InfiFPO: Implicit Model Fusion via Preference Optimization in Large Language Models
NeurIPS 2025 · Table Main Results
Alignment and preferences
481.7POISE m=2
Qwen3-4B
Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor’s Internal States
NeurIPS 2026 · Table 1
RL from feedback
578.29StressPrompt level 8
Llama-3-8B-Instruct
StressPrompt: Does Stress Impact Large Language Models and Human Performance Similarly?
AAAI 2025 · Table 1
AI and society
677.4UltraFeedback Best/Worst DPO
Llama-3.1-Tulu-3-8B-SFT
What Matters in Data for DPO?
NeurIPS 2025 · Table 1
Alignment and preferences
766.02Nemotron-Research-Reasoning-Qwen-1.5B
DeepSeek-R1-Distill-Qwen-1.5B
ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models
NeurIPS 2025 · Evaluation table, STEM and ins
RL from feedback
863.4512B 4-bit k-means QAT
Llama-style, same memory budget (~7.8GB weights)
1-Bit Wonder: Improving QAT Performance in the Low-Bit Regime through K-Means Quantization
ICML 2026 · Table 1
Quantization
963.22RAIN-Merging
DeepSeek-R1-Distill-Qwen-7B + Qwen2.5-7B-Instruct
RAIN-Merging: A Gradient-Free Method to Enhance Instruction Following in Large Reasoning Models with Preserved Thinking Format
ICLR 2026 · Table 1
Reasoning and chain of thought
1059.3Dream-Instruct + G-Star
Dream-Instruct 7B
Guided Star-Shaped Masked Diffusion
ICML 2026 · Table 1
Fast diffusion sampling
10 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

IFEval · accuracy

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

#accuracy ↑MethodPaperTopic
192.2IF-RLVR Tulu-DPO 8B
Tulu-DPO
Generalizing Verifiable Instruction Following
NeurIPS 2025 · Table: Model performances on d
Agents, code and math
288.8DMS 8×
Qwen3-8B
Inference-Time Hyper-Scaling with KV Cache Compression
NeurIPS 2025 · Table: DMS 8× Qwen3-8B
Attention and transformers
386.1HeRL
Qwen3-4B-Instruct-2507
Experience is the Best Teacher: Motivating Effective Exploration in Reinforcement Learning for LLMs
ICML 2026 · Table 1
RL from feedback
483.7PACE
Ministral-3-14B
PACE: Two-Timescale Self-Evolution for Small Language Model Agents
NeurIPS 2026 · Table 1
LLM agents
580.22InfiGFusion
Phi-4
InfiGFusion: Graph-on-Logits Distillation via Efficient Gromov-Wasserstein for Model Fusion
NeurIPS 2025 · Main results table
Reasoning and chain of thought
678.4LoongRL-14B
Qwen2.5-14B-Instruct
LoongRL: Reinforcement Learning for Advanced Reasoning over Long Contexts
ICLR 2026 · Table 2
Reasoning and chain of thought
776.1CGD
LLaMA3.1-8B Instruct
Critique-Guided Distillation for Robust Reasoning via Refinement
ICML 2026 · Table 2
Instruction and fine-tuning
871.35FLARE-9B (AR-Trust)
Qwen3.5-9B
FLARE: Diffusion for Hybrid Language Model
NeurIPS 2026 · Table 1
Language, safety and interpretability
969.5PROSPER
Qwen2.5-7B-Instruct
Back to Blackwell: Closing the Loop on Intransitivity in Multi-Objective Preference Fine-Tuning
NeurIPS 2026 · Table 2
Instruction and fine-tuning
1057.67PEAR+GRPO
Qwen3-0.6B-Base
Good SFT Optimizes for SFT, Better SFT Prepares for Reinforcement Learning
ICML 2026 · Appendix table
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

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