| QualityFlow: An Agentic Workflow for Program Synthesis Controlled by LLM Quality Checks | ✓ Link | 98.8 | QualityFlow (Sonnet-3.5) | 2025-01-20 |
| Planning-Driven Programming: A Large Language Model Programming Workflow | ✓ Link | 98.2 | Phi-2 | 2024-11-21 |
| NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness | ✓ Link | 98.17 | NeoHorse-1-9B | 2026-09-08 |
| Recursive Think-Answer Process for LLMs and VLMs | | 97.9 | Qwen-2.5-32B-R-TAP | 2026-03-02 |
| Introducing Claude Sonnet 4.5 | | 97.6 | Claude Sonnet 4.5 Thinking | 2025-09-29 |
| IQuest-Coder-V1 Technical Report | ✓ Link | 97.6 | IQuest-Coder-V1-40B-Loop-Instruct | 2026-03-17 |
| AgentConductor: Topology Evolution for Multi-Agent Competition-Level Code Generation | | 97.5 | AgentConductor (3B) | 2026-02-19 |
| DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning | ✓ Link | 97.4 | DeepSeek R1 | 2025-01-22 |
| Introducing Claude Opus 4.6 | | 97.0 | Claude Opus 4.6 Thinking | 2026-02-05 |
| GLM-5: From Vibe Coding to Agentic Engineering | ✓ Link | 97.0 | GLM 5 Thinking | 2026-02-11 |
| Execution Guided Line-by-Line Code Generation | ✓ Link | 96.95 | EG-CFG (DeepSeek-V3-0324) | 2025-06-12 |
| TPPO: Tracking-Value PPO For Language Model Reasoning | ✓ Link | 96.5 | TPPO (Qwen3.5-9B) | 2026-10-12 |
| CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoning | ✓ Link | 96.34 | CopT (Qwen3-8B) | 2026-05-19 |
| RAAS: LLM Agentic System Architecture Search with GRPO | ✓ Link | 96.31 | RAAS | 2026-06-03 |
| Introducing OpenAI o3 and o4-mini | | 96.3 | o4 Mini High | 2025-04-16 |
| Introspective Diffusion Language Models | ✓ Link | 96.3 | I-DLM-32B | 2026-04-13 |
| Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models | ✓ Link | 95.9 | Youtu-LLM 2B | 2025-12-31 |
| ReflexiCoder: Teaching Large Language Models to Self-Reflect on Generated Code and Self-Correct It via Reinforcement Learning | ✓ Link | 95.73 | ReflexiCoder-8B (Multiple) | 2026-03-06 |
| Claude Opus 4.1 | | 95.7 | Claude Opus 4.1 | 2025-08-05 |
| Unlocking Lossless Speedups in LLMs via Discrete Diffusion | ✓ Link | 95.2 | Uno (K2-Horizon-7B-Uno) | 2026-09-03 |
| Gemini 2.5: Our most intelligent AI model | | 95.1 | Gemini 2.5 Pro Preview 05-06 | 2025-03-25 |
| REAM: Merging Improves Pruning of Experts in LLMs | ✓ Link | 95.1 | REAM (Qwen3-Coder-Next, 384 experts, 0:0.7:0.3) | 2026-04-06 |
| FlowEvo: Self-Evolving Agents through the Co-Evolution of Workflows and Executable Skills | ✓ Link | 95.1 | FlowEvo | 2026-08-20 |
| Explore Broadly, Reason Sharply: Push Small Models toward the Frontier via Sampling | | 94.9 | Qwen3-8B | 2026-09-29 |
| LLaDA2.0: Scaling Up Diffusion Language Models to 100B | ✓ Link | 94.51 | LLaDA2.0-flash | 2025-12-10 |
| From Code to Correctness: Closing the Last Mile of Code Generation with Hierarchical Debugging | ✓ Link | 94.5 | DeepSeek-Coder-V2-Lite (MGDebugger) | 2024-10-02 |
| HyperCLOVA X THINK Technical Report | ✓ Link | 94.5 | HyperCLOVA X SEED 14B Think | 2025-06-27 |
| Introducing GPT-5.1 for developers | | 94.5 | GPT-5.1 | 2025-11-12 |
| MapCoder: Multi-Agent Code Generation for Competitive Problem Solving | ✓ Link | 93.9 | Mistral 7B | 2024-05-18 |
| DeepSeek-V3.2 Technical Report | ✓ Link | 93.9 | DeepSeek V3.2 Thinking | 2025-12-02 |
| Qwen3-Coder: Agentic Coding in the World | ✓ Link | 92.7 | Qwen3 Coder 480B A35B (exacto) | 2025-07-22 |
| Qwen3 Technical Report | ✓ Link | 92.1 | Qwen3 235B A22B Instruct 2507 | 2025-05-14 |
| ERNIE 4.5 Technical Report | ✓ Link | 92.1 | ERNIE-4.5-300B-A47B | 2025-06-30 |
| Introducing Claude 3.5 Sonnet | | 92.0 | Claude 3.5 Sonnet | 2024-06-20 |
| Atlas: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning | | 91.5 | Atlas (cluster) | 2026-01-07 |
| Think Anywhere in Code Generation | ✓ Link | 91.5 | Think-Anywhere (Qwen2.5-Coder-7B-Instruct) | 2026-03-31 |
| The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction | ✓ Link | 91.5 | Edge0-8B | 2026-09-16 |
| Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning | | 91.3 | ERC-DAPO (DeepSeek-R1-Distill-Qwen-7B) | 2025-12-05 |
| OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization | ✓ Link | 91.06 | OSCAR (GLM-4.7, 358B) | 2026-05-18 |
| Introducing Claude 3.5 Sonnet | | 90.85 | Claude Sonnet 3.5 | 2024-06-20 |
| From P(y|x) to P(y): Investigating Reinforcement Learning in Pre-train Space | ✓ Link | 90.24 | DSRL (Qwen3-8B) | 2026-04-15 |
| Safin-1: Safety from Within through Memory-Native State Evolution | ✓ Link | 90.24 | Safin-1-35B-A3B-SFT | 2026-08-31 |
| L2MAC: Large Language Model Automatic Computer for Extensive Code Generation | ✓ Link | 90.2 | L2MAC (GPT-4) | 2023-10-02 |
| Hello GPT-4o | | 90.2 | GPT-4o | 2024-05-13 |
| Mistral Large 3 | ✓ Link | 90.2 | Mistral Large 3 2512 | 2025-11-28 |
| TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning | | 89.6 | TERMINATOR (Ministral-3-14B-Reasoning-2512) | 2026-03-13 |
| Loop the Loopies! | | 89.02 | Loopie-20B-A2B (Thinking) | 2026-07-17 |
| Introducing Llama 3.1: Our most capable models to date | ✓ Link | 89.0 | Llama 3.1 (405B, Instruct) | 2024-07-23 |
| From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMs | ✓ Link | 89 | NBDiff-7B (Instruct) | 2025-12-07 |
| NGM: A Plug-and-Play Training-Free Memory Module for LLMs | ✓ Link | 88.41 | Qwen3-14B + NGM | 2026-05-16 |
| dots.llm1 Technical Report | ✓ Link | 88.4 | dots.llm1.inst | 2025-06-06 |
| Qwen2.5 Technical Report | | 87.8 | Qwen2.5-Plus | 2024-12-19 |
| Qwen2.5-VL Technical Report | ✓ Link | 87.8 | Qwen2.5-VL-72B | 2025-02-19 |
| Gemma 3 Technical Report | ✓ Link | 87.8 | Gemma 3 (27B, IT) | 2025-03-25 |
| GPT-4o mini: advancing cost-efficient intelligence | | 87.2 | GPT-4o mini | 2024-07-18 |
| DeepSeek-V3 Technical Report | ✓ Link | 87.2 | DeepSeek V3 | 2024-12-27 |
| Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance | ✓ Link | 87.2 | Falcon-H1-34B-Instruct | 2025-07-30 |
| Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding | | 87.2 | Nemotron-Labs-Diffusion-14B (Quad. SS) | 2026-07-07 |
| MiniMax-01: Scaling Foundation Models with Lightning Attention | ✓ Link | 86.9 | MiniMax-Text-01 | 2025-01-14 |
| Latent Reasoning with Normalizing Flows | | 86.7 | NF-CoT (Unified) + RL | 2026-06-04 |
| Stable-DiffCoder: Pushing the Frontier of Code Diffusion Large Language Model | ✓ Link | 86.6 | Stable-DiffCoder-8B-Instruct | 2026-01-22 |
| MiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal Interaction | ✓ Link | 86.6 | MiniCPM-o 4.5-Instruct | 2026-04-30 |
| Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-Training | ✓ Link | 86.6 | Qwen3-14B with GCWM | 2026-05-10 |
| Qwen2 Technical Report | ✓ Link | 86.0 | Qwen2-72B-Instruct | 2024-07-15 |
| MixSD: Mixed Contextual Self-Distillation for Knowledge Injection | | 86 | Qwen3-8B with MixSD (lambda=0.3) | 2026-05-16 |
| Introducing the next generation of Claude | | 84.9 | Claude 3 Opus | 2024-03-04 |
| The Llama 4 herd | ✓ Link | 84.8 | Llama 4 Maverick | 2025-04-05 |
| ACES: Who Tests the Tests? Leave-One-Out AUC Consistency for Code Generation | | 84.15 | ACES-O | 2026-04-05 |
| Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context | | 84.1 | Gemini 1.5 Pro | 2024-03-08 |
| DarkForest: Less Talk, Higher Accuracy for Multi-Agent LLMs | ✓ Link | 84 | DarkForest | 2026-05-24 |
| Fast and Accurate Causal Parallel Decoding using Jacobi Forcing | ✓ Link | 83.5 | Jacobi Forcing Model (MR) | 2025-12-16 |
| BatCoder: Self-Supervised Bidirectional Code-Documentation Learning via Back-Translation | | 83.5 | BatCoder 7B | 2026-01-30 |
| UniSD: Towards a Unified Self-Distillation Framework for Large Language Models | ✓ Link | 83.5 | UniSD* | 2026-05-07 |
| Debug like a Human: A Large Language Model Debugger via Verifying Runtime Execution Step-by-step | ✓ Link | 82.9 | LLaMA 3 | 2024-02-25 |
| Command A: An Enterprise-Ready Large Language Model | ✓ Link | 82.9 | Command A | 2025-03-13 |
| CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM Agents | ✓ Link | 82.5 | CERA-MoA (Qwen3-4B) | 2026-09-16 |
| Reading, Not Thinking: Understanding and Bridging the Modality Gap When Text Becomes Pixels in Multimodal LLMs | | 82.32 | Qwen3-VL-8B (LM-only, self-distilled) | 2026-03-10 |
| InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models | ✓ Link | 82.3 | InternVL3-78B | 2025-04-14 |
| ThinkRouter: Efficient Reasoning via Routing Thinking between Latent and Discrete Spaces | | 81.82 | ThinkRouter (Qwen3-1.7B) | 2026-02-12 |
| MARS: Enabling Autoregressive Models Multi-Token Generation | ✓ Link | 81.7 | MARS-7B (B=4) | 2026-04-08 |
| Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models | | 81.7 | MELT-1.6B | 2026-05-08 |
| DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model | ✓ Link | 81.1 | DeepSeek-V2 Chat (RL) (0-shot) | 2024-05-07 |
| The Llama 4 herd | ✓ Link | 81.1 | Llama 4 Scout | 2025-04-05 |
| Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters | ✓ Link | 81.1 | Step-3.5-Flash Base | 2026-02-11 |
| Introducing Llama 3.1: Our most capable models to date | ✓ Link | 80.5 | Llama 3.1 (70B, Instruct) | 2024-07-23 |
| Qwen2 Technical Report | ✓ Link | 79.9 | Qwen2-57B-A14B-Instruct | 2024-07-15 |
| Large Flow Language Models | ✓ Link | 79.9 | Lafa 8B | 2026-10-20 |
| Decoding Looped Transformers Better for (Almost) Free | | 79.88 | LoopCD-Ouro-2.6B (adaptive) | 2026-10-01 |
| SAHOO: Safeguarded Alignment for High-Order Optimization Objectives in Recursive Self-Improvement | ✓ Link | 79.5 | SAHOO (Qwen3-8B) | 2026-03-06 |
| Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning | ✓ Link | 79.4 | Nemotron 3 Super (120B-A12B Base) | 2026-04-14 |
| DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression | ✓ Link | 79.4 | DeepSeek-V4.1-Flash Base | 2026-09-17 |
| Hybrid Policy Distillation for LLMs | ✓ Link | 79.3 | HPD (Qwen2.5-Coder, 1.5B) | 2026-04-22 |
| Prism: Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models | ✓ Link | 79.27 | Prism (LLaDA-8B-Instruct, K=8) | 2026-02-02 |
| Qwen2.5-Omni Technical Report | ✓ Link | 78.7 | Qwen2.5-Omni-7B | 2025-03-26 |
| S2D2: Fast Decoding for Diffusion LLMs via Training-Free Self-Speculation | ✓ Link | 78.7 | S2D2 (SDAR-8B-Chat, config-A) | 2026-03-26 |
| S2D2: Fast Decoding for Diffusion LLMs via Training-Free Self-Speculation | ✓ Link | 78.7 | S2D2 (SDAR-8B-Chat) | 2026-03-26 |
| ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding | ✓ Link | 78.66 | ReFusion | 2025-12-15 |
| ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools | ✓ Link | 78.5 | GLM-4 (0520) | 2024-06-18 |
| InCoder-32B: Code Foundation Model for Industrial Scenarios | ✓ Link | 78.3 | InCoder-32B (32B) | 2026-03-17 |
| Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning | ✓ Link | 78.05 | Nemotron 3 Nano 30B-A3B Base | 2025-12-23 |
| CharacterFlywheel: Scaling Iterative Improvement of Engaging and Steerable LLMs in Production | | 77.4 | CharacterFlywheel V7 (Llama 3.1 70B) | 2026-03-02 |
| LoopRPT: Reinforcement Pre-Training for Looped Language Models | | 76.83 | Ouro (2.6B) LoopRPT | 2026-03-20 |
| T5Gemma 2: Seeing, Reading, and Understanding Longer | | 76.8 | T5Gemma 2 4B-4B (post-trained) | 2025-12-16 |
| daVinci-Dev: Agent-native Mid-training for Software Engineering | ✓ Link | 76.73 | daVinci-Dev-72B (MT Mix) | 2026-01-26 |
| daVinci-Env: Open SWE Environment Synthesis at Scale | ✓ Link | 76.25 | OpenSWE-72B (Qwen2.5-72B-Base) | 2026-03-13 |
| LFPO: Likelihood-Free Policy Optimization for Masked Diffusion Models | ✓ Link | 75.6 | LFPO (All Loss) | 2026-03-02 |
| Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models | | 74.5 | IMH (WeDLM-8B) | 2026-04-01 |
| Gemini: A Family of Highly Capable Multimodal Models | ✓ Link | 74.4 | Gemini Ultra | 2023-12-19 |
| Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs | ✓ Link | 74.4 | Phi-4-Mini | 2025-02-24 |
| DeepSeek LLM: Scaling Open-Source Language Models with Longtermism | ✓ Link | 73.8 | DeepSeek LLM 67B Chat | 2024-01-05 |
| A Sovereign, Open-Source Foundation Model for German and English | ✓ Link | 73.8 | Soofi S 30B-A3B Base | 2026-07-10 |
| Motif 3: Technical Report | ✓ Link | 73.7 | Motif 3 Base (zero-shot) | 2026-08-10 |
| Nemotron-4 340B Technical Report | ✓ Link | 73.2 | Nemotron-4-340B-Instruct | 2024-06-17 |
| Aria: An Open Multimodal Native Mixture-of-Experts Model | ✓ Link | 73.2 | Aria | 2024-10-08 |
| Deep Fusion Memory: A Chunk-Level Parametric Memory for Large Language Models | | 73.02 | DFM (Qwen3-8B-Base) | 2026-10-18 |
| The Llama 3 Herd of Models | ✓ Link | 72.6 | Llama 3.1 (8B, Instruct) | 2024-07-31 |
| Who Should Lead Decoding Now? Tracking Reliable Trajectories for Ensembling Masked Diffusion Language Models | | 72.56 | TIE (DreamCoder + DiffuCoder; Token change count) | 2026-06-15 |
| Pixtral 12B | ✓ Link | 72 | Pixtral 12B | 2024-10-09 |
| Encoder-Decoder Gemma: Improving the Quality-Efficiency Trade-Off via Adaptation | ✓ Link | 72 | T5Gemma 9B-9B PrefixLM IT | 2025-04-08 |
| Every Token Counts: Generalizing 16M Ultra-Long Context in Large Language Models | ✓ Link | 71.95 | HSA-UltraLong-Inst | 2025-11-28 |
| Claude 2 | | 71.2 | Claude 2 | 2023-07-11 |
| Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone | ✓ Link | 70.7 | Phi-3.5-MoE | 2024-04-22 |
| Group Sequence Policy Optimization | | 69.3 | AliceAI-T5-35B-A0.6B-Base | 2025-07-24 |
| Scalable Prompt Routing via Fine-Grained Latent Task Discovery | | 68.4 | FineRouter | 2026-03-19 |
| Co-RL: Unsupervised Reasoning Emerges from Diverse Cohort in Multi-agent RL | ✓ Link | 68.3 | Co-RL (Llama-3.1-8B-Instruct, Same family) | 2026-08-19 |
| Removing the NEEDLE in the Haystack: Backdoor Removal in LLMs via Weight Orthogonalisation | ✓ Link | 68.29 | NEEDLE (Qwen3-4B-Instruct-2507, Targeted refusal, Sleeper) | 2026-09-29 |
| Code Llama: Open Foundation Models for Code | ✓ Link | 67.8 | Code Llama - Instruct 70B | 2023-08-24 |
| Grammar-Constrained Decoding Can Jailbreak LLMs into Generating Malicious Code | ✓ Link | 67.48 | CodeShield (Qwen2.5-Coder-7B) | 2026-06-10 |
| GPT-4 Technical Report | ✓ Link | 67 | GPT-4 (0-shot) | 2023-03-15 |
| Improved Large Language Diffusion Models | ✓ Link | 65.9 | iLLaDA 8B Instruct | 2026-06-24 |
| MobileMoE: Scaling On-Device Mixture of Experts | | 65.2 | MobileMoE-L (922M) | 2026-05-26 |
| Introducing LFM-7B: Setting New Standards for Efficient Language Models | | 63.41 | LFM-7B | 2025-01-20 |
| Token-Level LLM Collaboration via FusionRoute | | 63 | FusionRoute (Llama-3.1-8B) | 2026-01-08 |
| Attention Residuals | ✓ Link | 62.2 | Kimi Linear (48B) with Block AttnRes | 2026-03-16 |
| daVinci-LLM:Towards the Science of Pretraining | ✓ Link | 61.64 | daVinci-LLM-3B | 2026-03-28 |
| Nemotron-H: A Family of Accurate and Efficient Hybrid Mamba-Transformer Models | ✓ Link | 61 | Nemotron-H-47B-Base | 2025-04-04 |
| SoloQ: Calibration-Free Quantization for Diffusion Language Models | | 59.1 | SoloQ-N (Fast-dLLM v2-7B) | 2026-10-05 |
| Nemotron-H: A Family of Accurate and Efficient Hybrid Mamba-Transformer Models | ✓ Link | 58.5 | Nemotron-H-8B-Base | 2025-04-04 |
| Dream 7B: Diffusion Large Language Models | | 57.9 | Dream 7B | 2025-08-21 |
| Rethinking Token Reweighting for SFT: Suppress, Reverse, and Extrapolate Learned Features | | 57.11 | SCALE (Qwen2.5-Math-7B) | 2026-09-27 |
| dOPSD: On-Policy Self-Distillation for Diffusion Language Models | ✓ Link | 56.71 | Dream-7B-Instruct (dOPSD) | 2026-07-05 |
| DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data | ✓ Link | 56.7 | Mimir 1B | 2026-08-13 |
| Let's Scale Step by Step: Compute-Efficient Hyperparameter Transfer for Large-Scale Mixture-of-Experts | | 56.3 | Foundation MoE (155B total, 17B active) | 2026-08-20 |
| StepAudio 3 Gen Technical Report | | 54.27 | StepAudio 3 Gen (pretrained) | 2026-09-11 |
| Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation | ✓ Link | 53.05 | Pion (Qwen2.5-1.5B finetuning) | 2026-05-12 |
| Mid-Training with Self-Generated Data Improves Reinforcement Learning in Language Models | | 52.82 | Llama 3.2–3B–Instruct (mid-trained, n=64, RL) | 2026-05-08 |
| Gemma 2: Improving Open Language Models at a Practical Size | ✓ Link | 51.8 | Gemma 2 27B | 2024-07-31 |
| Textbooks Are All You Need | ✓ Link | 50.6 | Phi-1 | 2023-06-20 |
| Improved Large Language Diffusion Models | ✓ Link | 50 | iLLaDA 8B Base | 2026-06-24 |
| LLaDA MoE v2: Scaling Mixture-of-Experts Diffusion Language Models | | 50 | LLaDA MoE v2 Base | 2026-08-04 |
| Large Language Diffusion Models | ✓ Link | 49.4 | LLaDA 8B Instruct | 2025-02-14 |
| The Flexibility Trap: Why Arbitrary Order Limits Reasoning Potential in Diffusion Language Models | ✓ Link | 49.4 | JustGRPO (LLaDA 8B Instruct) | 2026-01-21 |
| Turning the TIDE: Cross-Architecture Distillation for Diffusion Large Language Models | ✓ Link | 49.39 | Qwen3-0.6B-BD3LM (Tide-Shared) | 2026-04-29 |
| dLLM: Simple Diffusion Language Modeling | ✓ Link | 46.3 | Qwen3-0.6B-diffusion-bd3lm-v0.1 | 2026-02-26 |
| Stepped MoE: Segment-Level Routing with Configurable Inference Complexity | | 46.1 | Flexible S-MoE 12B→4B (extra data) | 2026-10-05 |
| What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models | | 43.9 | OnTrack (Ours) | 2026-10-22 |
| Puro-2B: Poor Lab's Qwen2-1.5B Trained on RTX 5090 within $5090 | ✓ Link | 42.27 | Puro-2B UD Tulu3 SFT | 2026-08-27 |
| Less is More: Early Stopping Rollout for On-Policy Distillation | | 42.1 | Qwen2.5-Math-1.5B (ESR) | 2026-05-26 |
| Normalized Low-Rank Adaptation | ✓ Link | 42.1 | Llama3.2-3B NoRA | 2026-08-31 |
| Mellum2 Technical Report | ✓ Link | 41.5 | Mellum 2 (base) | 2026-05-29 |
| Decentralized Instruction Tuning: Conflict-Aware Splitting and Weight Merging | ✓ Link | 41.5 | MERIT (Proposed, 2D split, 4 groups, Qwen2.5-3B) | 2026-06-01 |
| Stop the Flip-Flop: Context-Preserving Verification for Fast Revocable Diffusion Decoding | | 41.46 | COVER (LLaDA-Instruct-8B) | 2026-02-05 |
| Flash-dLLM: IO-Aware KV Caching and Parallel Decoding for Fast, Memory-Efficient Diffusion LLMs | ✓ Link | 40.24 | Flash-dLLM | 2026-09-22 |
| Mixtral of Experts | ✓ Link | 40.2 | Mixtral 8x7B | 2024-01-08 |
| Encoder-Decoder Gemma: Improving the Quality-Efficiency Trade-Off via Adaptation | ✓ Link | 40.2 | T5Gemma 9B-9B PrefixLM | 2025-04-08 |
| MARS: Enabling Autoregressive Models Multi-Token Generation | ✓ Link | 40.2 | MARS (Qwen2.5-0.5B-Instruct) | 2026-04-08 |
| Understanding by Reconstruction: Reversing the Software Development Process for LLM Pretraining | | 37.2 | Llama3-8B-Instruct (Repo2Agent-Search) | 2026-03-11 |
| The Falcon Series of Open Language Models | ✓ Link | 35.4 | Falcon-180B | 2023-11-28 |
| Byte Latent Transformer: Patches Scale Better Than Tokens | ✓ Link | 35.4 | BLT-Entropy 8B | 2024-12-13 |
| ReMix: Reinforcement routing for mixtures of LoRAs in LLM finetuning | | 32.93 | ReMix (Llama 3 8B) | 2026-03-10 |
| Gemma: Open Models Based on Gemini Research and Technology | ✓ Link | 32.3 | Gemma 7B | 2024-03-13 |
| Nemotron-4 15B Technical Report | ✓ Link | 31.6 | Nemotron-4 15B | 2024-02-26 |
| Compact Language Models via Pruning and Knowledge Distillation | ✓ Link | 31.6 | Minitron 8B | 2024-07-19 |
| RecurrentGemma: Moving Past Transformers for Efficient Open Language Models | ✓ Link | 31.1 | RecurrentGemma-9B | 2024-04-11 |
| Mistral 7B | ✓ Link | 30.5 | Mistral 7B | 2023-10-10 |
| Register Tokens for Bounded-State Reasoning in Diffusion Language Models | ✓ Link | 30.5 | Dream-7B-Base + Registers | 2026-09-14 |
| Llama 2: Open Foundation and Fine-Tuned Chat Models | ✓ Link | 29.9 | Llama 2 70B | 2023-07-18 |
| Falcon2-11B Technical Report | | 29.6 | Falcon2-11B | 2024-07-20 |
| Jamba: A Hybrid Transformer-Mamba Language Model | ✓ Link | 29.3 | Jamba | 2024-03-28 |
| CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis | ✓ Link | 29.28 | CodeGen-Mono 16.1B | 2022-03-25 |
| SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model | ✓ Link | 28.1 | SmolLM2 1.7B Instruct | 2025-02-04 |
| Power Scheduler: A Batch Size and Token Number Agnostic Learning Rate Scheduler | ✓ Link | 26.8 | PowerLM-3B | 2024-08-23 |
| PaLM: Scaling Language Modeling with Pathways | ✓ Link | 26.2 | PaLM 540B | 2022-04-05 |
| LLaMA: Open and Efficient Foundation Language Models | ✓ Link | 23.7 | LLaMA-65B | 2023-02-27 |
| Compact Language Models via Pruning and Knowledge Distillation | ✓ Link | 23.3 | Minitron 4B | 2024-07-19 |
| SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model | ✓ Link | 22.6 | SmolLM2 1.7B | 2025-02-04 |
| Sumi: Open Uniform Diffusion Language Model from Scratch | ✓ Link | 22.6 | Sumi-7B | 2026-06-17 |
| Fast Byte Latent Transformer | | 22.56 | BLT-S (3B, k=4) | 2026-05-08 |
| Copy-Paste Language Models | ✓ Link | 22.1 | Copy-Paste Language Model (1.7B, Hybrid 3:1) | 2026-10-13 |
| RecurrentGemma: Moving Past Transformers for Efficient Open Language Models | ✓ Link | 21.3 | RecurrentGemma-2B | 2024-04-11 |
| SmolLM - blazingly fast and remarkably powerful | ✓ Link | 20.76 | SmolLM 1.7B | 2024-07-16 |
| Releasing Persimmon-8B | ✓ Link | 20.7 | Persimmon-8B-FT | 2023-09-07 |
| SantaCoder: don't reach for the stars! | ✓ Link | 18 | SantaCoder | 2023-01-09 |
| BLOOM: A 176B-Parameter Open-Access Multilingual Language Model | ✓ Link | 15.52 | BLOOM-176B | 2022-11-09 |
| Universal YOCO for Efficient Depth Scaling | | 10.37 | YOCO-U (Equal Steps) | 2026-04-01 |
| Scaling Inherently Interpretable Language Models | ✓ Link | 8.5 | Steerling-8B | 2026-08-06 |
| Training Language Models via Neural Cellular Automata | ✓ Link | 7.5 | NCA-pretrained 1.6B | 2026-03-09 |
| ModuleFormer: Modularity Emerges from Mixture-of-Experts | ✓ Link | 5.51 | MoLM-8B-K2 | 2023-06-07 |