| Mach-Mind-4-Flash Technical Report | | 82.82 | Mach-Mind-4-Flash | 2026-07-10 |
| Inkling-Small | ✓ Link | 82.2 | Inkling-Small | 2026-07-27 |
| Sage-A1.5: From Long-Horizon Execution to Scientific Exploration with Capability Composition | ✓ Link | 81.6 | Sage-A1.5 | 2026-10-20 |
| Qwen3.8-Flash-Next | ✓ Link | 81.3 | Qwen3.8-Flash-Next | 2026-08-24 |
| Group Entropy-Controlled Policy Optimization | | 80.7 | GEPO (Qwen3.5-9B) | 2026-07-18 |
| Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent | ✓ Link | 80.6 | Agents-A1 | 2026-06-29 |
| SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture | ✓ Link | 79.79 | SenseNova-U1-A3B-MoT | 2026-05-12 |
| Introducing Beam: Reflection's 501B open-weight model | | 79.7 | Beam | 2026-10-05 |
| Qwen3.8-27B | ✓ Link | 79.5 | Qwen3.8-27B | 2026-08-05 |
| Closing the Loop: Practical Training Recipes for Looped Language Models | | 79.0 | Dory1.0-4B | 2026-09-30 |
| Motif 3: Technical Report | ✓ Link | 78.2 | Motif 3 | 2026-08-10 |
| MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training | | 77.89 | Qwen3-30B-A3B + MOPD | 2026-06-29 |
| Unified Audio Intelligence Without Regressing on Text Intelligence | | 77.8 | Audex 30B-A3B | 2026-07-06 |
| Muse Glimmer 30B: An On-Device Agentic Model | ✓ Link | 77.0 | Muse Glimmer-30B (High Reasoning) | 2026-10-17 |
| Rufus-Air: An Open LLM Post-Training Recipe | | 76.9 | Rufus-Air | 2026-09-24 |
| Qwen3.5: Towards Native Multimodal Agents | ✓ Link | 76.5 | Qwen3.5-27B | 2026-02-16 |
| Qwen3.5: Towards Native Multimodal Agents | ✓ Link | 76.5 | Qwen3.5-397B-A17B | 2026-02-16 |
| Qwen3.5: Towards Native Multimodal Agents | ✓ Link | 76.1 | Qwen3.5-122B-A10B | 2026-02-16 |
| AgentGrad: Intervention-guided Prompt Optimization for Multi Agent Systems | | 76.08 | AgentGrad (GPT-5-mini) | 2026-09-08 |
| The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence | ✓ Link | 76.0 | MiniMax-M2.7 | 2026-05-26 |
| LFM2.5-24B-A2B: Built for Fast On-Device Agents | | 74.69 | LFM2.5-24B-A2B | 2026-10-13 |
| VibeThinker-3B: Exploring the Frontier of Verifiable Reasoning in Small Language Models | ✓ Link | 74.5 | VibeThinker-3B | 2026-06-15 |
| Nemotron 3 Nano Omni: Efficient and Open Multimodal Intelligence | ✓ Link | 74.2 | Nemotron 3 Nano Omni | 2026-04-27 |
| RUBRIC-ARROW: Alternating Pointwise Rubric Reward Modeling for LLM Post-training in Non-verifiable Domains | ✓ Link | 73.2 | RUBRIC-ARROW-voting@5 | 2026-05-27 |
| NVIDIA Nemotron 3: Efficient and Open Intelligence | ✓ Link | 72.56 | Nemotron 3 Super | 2025-12-24 |
| One Model to Loop Them All: Elastic Latent Recurrence for Language Models | | 71.8 | Koi-4B | 2026-10-08 |
| Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning | ✓ Link | 71.5 | NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 | 2025-12-23 |
| Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale | | 71.2 | Intern-S1-Pro | 2026-03-26 |
| Qwen3.5: Towards Native Multimodal Agents | ✓ Link | 70.2 | Qwen3.5-35B-A3B | 2026-02-16 |
| Vela: User Behavior as Supervision Sparks LLM Intelligence | ✓ Link | 69.2 | Vela-35B-A3B (extra data) | 2026-10-10 |
| MAI-Thinking-1: Building a Hill-Climbing Machine | | 69 | MAI-Thinking-1 | 2026-06-02 |
| SenseNova-U1.5: Towards Native Unified Visual Intelligence | ✓ Link | 69.0 | SenseNova-U1.5-8B-MoT | 2026-09-10 |
| Olmo 3 | ✓ Link | 68.1 | Olmo 3.1 Think 32B | 2025-12-15 |
| Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters | ✓ Link | 67.4 | Step-3.5-Flash | 2026-02-11 |
| K-EXAONE Technical Report | ✓ Link | 67.3 | K-EXAONE (236B, reasoning) | 2026-01-05 |
| KAT-Coder-V2 Technical Report | | 67.0 | KAT-Coder-V2 | 2026-03-29 |
| NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness | ✓ Link | 66.33 | NeoHorse-1-9B | 2026-09-08 |
| Data Science and Technology Towards AGI Part I: Tiered Data Management | ✓ Link | 66.3 | MiniCPM5-2B | 2026-02-09 |
| Qwen3.5: Towards Native Multimodal Agents | ✓ Link | 64.5 | Qwen3.5-9B | 2026-02-16 |
| EXAONE 4.5 Technical Report | ✓ Link | 62.6 | EXAONE 4.5 (Reasoning) | 2026-04-09 |
| LFM2.5-2.6B: Deploy Agents Everywhere | ✓ Link | 59.17 | LFM2.5-2.6B | 2026-08-04 |
| Announcing Bonsai 27B: The First 27B-Class Model to Run on a Phone | ✓ Link | 58.50 | Ternary Bonsai 27B | 2026-07-14 |
| Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale | ✓ Link | 57.62 | Ling-2.6-1T | 2026-06-13 |
| Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale | ✓ Link | 57.40 | Ling-2.6-flash | 2026-06-13 |
| Ling-2.6-1T: A Trillion-Parameter Comprehensive Flagship Model for Complex Tasks | | 57.00 | Ling-2.6-1T | 2026-04-29 |
| Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters | ✓ Link | 56.8 | Step-3.5-Flash PaCoRe | 2026-02-11 |
| LFM2.5-8B-A1B: an Even Better on-Device Mixture-of-Experts | ✓ Link | 56.47 | LFM2.5-8B-A1B | 2026-05-28 |
| Nanbeige4.2-3B: Unlocking Agentic Capabilities in a Compact Model | ✓ Link | 54.6 | Nanbeige4.2-3B | 2026-07-21 |
| Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling | ✓ Link | 53.4 | Falcon-H1R (7B) | 2026-01-05 |
| Qwen3.5-Omni Technical Report | | 52.6 | Qwen3.5-Omni-Plus | 2026-04-21 |
| Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy Distillation | | 51.72 | TGOPD (4B multi-domain GRPO-fallback) | 2026-09-02 |
| RubricHub: A Comprehensive and Highly Discriminative Rubric Dataset via Automated Coarse-to-Fine Generation | ✓ Link | 51.4 | Qwen3-14B (RuFT→RuRL) | 2026-01-13 |
| Shepherd: A Runtime Substrate Empowering Meta-Agents with a Formalized Execution Trace | | 51.3 | CRO | 2026-05-11 |
| HY-MT1.5 Technical Report | ✓ Link | 50.67 | Hy-MT2-30B-A3B | 2025-12-30 |
| ROSS: Relearning from Self-Generated Rollouts through Selective Supervision | | 47.76 | ROSS (Qwen3.6-35B-A3B) | 2026-09-28 |
| Introducing LFM2.5: The Next Generation of On-Device AI | | 47.33 | LFM2.5-1.2B-Instruct | 2026-01-05 |
| Post-Trained MoE Can Skip Half Experts via Self-Distillation | ✓ Link | 47.3 | ZEDA (GLM-4.7-Flash) | 2026-05-18 |
| MiniCPM5-1B | ✓ Link | 46.67 | MiniCPM5-1B (Thinking) | 2026-05-25 |
| LFM2.5-1.2B-Thinking: On-Device Reasoning Under 1GB | | 44.85 | LFM2.5-1.2B-Thinking | 2026-01-20 |
| HarnessRL: Co-Evolving Policies with a Just-in-Time Training Harness for Open-Ended RL | ✓ Link | 43.6 | HarnessRL (Qwen3.5-9B) | 2026-10-19 |
| Trust Region On-Policy Distillation | | 42.18 | TrOPD (Qwen3-SFT-1.7B) | 2026-05-31 |
| CoRT: Counterfactual Replay for Token-Level Rubric-Guided Policy Optimization | | 41.09 | CoRT (Qwen3-14B) | 2026-07-28 |
| LFM2.5-350M: No Size Left Behind | | 40.69 | LFM2.5-350M | 2026-03-31 |
| Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation | ✓ Link | 38.7 | DN-MOPD-9B | 2026-09-28 |
| LFM2.5-230M: Built to Run Anywhere | | 38.40 | LFM2.5-230M | 2026-06-25 |
| Bootstrapping Exploration with Group-Level Natural Language Feedback in Reinforcement Learning | ✓ Link | 38.33 | GOLF (Qwen-3-8B) | 2026-03-04 |
| p1: Better Prompt Optimization with Fewer Prompts | | 37.41 | p1 p1 (Qwen3-4B-Instruct-2507) | 2026-04-09 |
| Cosmos 3: Omnimodal World Models for Physical AI | ✓ Link | 37.0 | Cosmos3-Super | 2026-06-01 |
| ACLArena: Agent Continue Learning in Multi-stage Post-training | ✓ Link | 29.1 | MLE (ACLArena) (extra data) | 2026-09-21 |
| MOPD-Router: Rethinking Teacher Routing in Multi-Teacher On-Policy Distillation | ✓ Link | 28.91 | MOPD-Router (ExpertAlign, Qwen3-4B-Non-Thinking student, domain-labeled) | 2026-09-25 |
| Enhancing Rubric-based RL via Self-Distillation | | 28.3 | CriPO-RS (Qwen3-4B) | 2026-07-21 |
| Weak-to-Strong On-Policy Distillation | ✓ Link | 27.0 | W2S-OPD (pre-/post-RL contrast) | 2026-07-28 |
| LFM2 Technical Report | ✓ Link | 25.85 | LFM2-8B-A1B | 2025-11-28 |
| Open-MOPD: Diagnosing and Fixing Capability Imbalance in Multi-Teacher On-Policy Distillation | ✓ Link | 24.67 | Open-MOPD (SmolLM3-3B) | 2026-08-19 |
| MobileMoE: Scaling On-Device Mixture of Experts | | 20.1 | MobileMoE-L (922M) | 2026-05-26 |
| ARC: Fair Relative Advantage Comparison in Open-Ended Real-World Interaction | | 17.35 | DAPO + ARC (Qwen3-8B) | 2026-08-13 |