| DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression | ✓ Link | 100.0 | DeepSeek-V4.1-Flash Max | 2026-09-17 |
| GLM-5.2: Built for Long-Horizon Tasks | ✓ Link | 99.2 | GLM-5.2 | 2026-06-16 |
| Introducing Beam: Reflection's 501B open-weight model | | 97.8 | Beam | 2026-10-05 |
| Inkling: Our open-weights model | ✓ Link | 97.1 | Inkling | 2026-07-15 |
| Kolibri: A Sovereign European Model on the Pareto Frontier | ✓ Link | 96.0 | Kolibri | 2026-10-03 |
| Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale | ✓ Link | 95.78 | Ring-2.6-1T (xhigh) | 2026-06-13 |
| Inkling-Small | ✓ Link | 95.5 | Inkling-Small | 2026-07-27 |
| Intern-S2-Mobius: Foundation Model with Decoupled Knowledge and Reasoning | ✓ Link | 95.31 | Intern-S2-Mobius-35B | 2026-08-14 |
| GLM-5.1: Towards Long-Horizon Tasks | ✓ Link | 95.3 | GLM-5.1 | 2026-04-03 |
| Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation | | 95.0 | Nemotron-Cascade-2-30B-A3B | 2026-03-19 |
| Muse Glimmer 30B: An On-Device Agentic Model | ✓ Link | 94.7 | Muse Glimmer-30B (High Reasoning) | 2026-10-17 |
| MAI-Thinking-1: Building a Hill-Climbing Machine | | 94.5 | MAI-Thinking-1 | 2026-06-02 |
| The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence | ✓ Link | 94.2 | MiniMax-M2.7 | 2026-05-26 |
| Qwen3.6 | ✓ Link | 94.1 | Qwen3.6-27B | 2026-04-21 |
| Limite 1B - Violetto: Technical Report | ✓ Link | 94.01 | Limite 1B - Violetto | 2026-10-06 |
| A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM | ✓ Link | 93.3 | A*-Thought-V2 (Qwen3.6-27B, tau=90) | 2026-09-07 |
| Grounded Skill Synthesis from Code at Scale for Agentic Intelligence | ✓ Link | 93.22 | GPT 5.2 + Code2Skill (reasoning) | 2026-09-04 |
| Qwen3.6 | ✓ Link | 92.7 | Qwen3.6-35B-A3B | 2026-04-21 |
| Mach-Mind-4-Flash Technical Report | | 92.70 | Mach-Mind-4-Flash | 2026-07-10 |
| EXAONE 4.5 Technical Report | ✓ Link | 92.6 | EXAONE 4.5 (Reasoning) | 2026-04-09 |
| Vela: User Behavior as Supervision Sparks LLM Intelligence | ✓ Link | 92.6 | Vela-35B-A3B (extra data) | 2026-10-10 |
| AgentMercury: Your Agent Can Synthesize Verifiable Environments for Business Scenarios at scale | | 92.2 | Qwen3.5-35B-A3B + SAO + Ours | 2026-08-21 |
| Thought-Level Beam Search for Reasoning | ✓ Link | 90.0 | Gambit (Phi-4-reasoning-plus-14B) | 2026-08-11 |
| Thomson: Continual Learning of Frontier Models for SovereignAI | | 90.0 | Thomson-1.0-Small | 2026-08-24 |
| Learning More by Training Less: Uncertainty-Guided Progressive Contraction for Efficient Reasoning Post-Training | ✓ Link | 90.0 | Qwen3.5-9B (rho=30%) | 2026-10-09 |
| Streaming Communication in Multi-Agent Reasoning | ✓ Link | 89.58 | STREAM-MA (Claude Opus 4.6, Chain) | 2026-06-03 |
| Unified Audio Intelligence Without Regressing on Text Intelligence | | 89.4 | Audex 30B-A3B | 2026-07-06 |
| Gemma 4 Technical Report | ✓ Link | 89.2 | Gemma 4 31B | 2026-07-02 |
| Announcing Bonsai 27B: The First 27B-Class Model to Run on a Phone | ✓ Link | 87.50 | Ternary Bonsai 27B | 2026-07-14 |
| Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale | ✓ Link | 87.40 | Ling-2.6-1T | 2026-06-13 |
| STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantization | ✓ Link | 87.24 | Qwen3.8-27B STEPQuant@6bit | 2026-09-29 |
| Recursive Multi-Agent Systems | ✓ Link | 86.7 | RecursiveMAS | 2026-04-28 |
| ParaTempo: Efficient Parallel Reasoning via Temporal Confidence | ✓ Link | 86.7 | ParaTempo (GPT-OSS-20B) | 2026-08-17 |
| Rufus-Air: An Open LLM Post-Training Recipe | | 86.7 | Rufus-Air | 2026-09-24 |
| The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction | ✓ Link | 86.6 | Edge0-35B-A3B-preview | 2026-09-16 |
| MiniCPM5-2B | ✓ Link | 86.5 | MiniCPM5-2B | 2026-09-07 |
| Unlocking Lossless Speedups in LLMs via Discrete Diffusion | ✓ Link | 86.3 | Uno (K2-Horizon-7B-Uno) | 2026-09-03 |
| PACT: From Credit Assignment to Critic Alignment | ✓ Link | 85.21 | PACT (Qwen3.5-4B) | 2026-09-22 |
| Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy Distillation | | 78.7 | TGOPD (35B single-domain GRPO-fallback) | 2026-09-02 |
| Gemma 4 Technical Report | ✓ Link | 77.5 | Gemma 4 12B Unified | 2026-07-02 |
| Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information | ✓ Link | 76.1 | Olmo3-7B-TK + AntiSD | 2026-05-12 |
| ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search | ✓ Link | 75.0 | ZGCM-1 | 2026-09-11 |
| Post-Trained MoE Can Skip Half Experts via Self-Distillation | ✓ Link | 74.4 | ZEDA (GLM-4.7-Flash) | 2026-05-18 |
| SPADE: Self-Play in Adaptive Synthetic Executable Environments | ✓ Link | 74.4 | SPADE (Games, Qwen3-30B-A3B-Instruct-2507) | 2026-08-19 |
| TTPO: Test-Time Policy Optimization | ✓ Link | 74.2 | Qwen3-8B + TTPO | 2026-08-27 |
| Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale | ✓ Link | 73.85 | Ling-2.6-flash | 2026-06-13 |
| Thinking Outside the Box: Can Language Models Rely on External Guidance Selectively? | | 73.3 | Thinking Outside the Box + SFT + RL_env (Qwen3-4B) (extra data) | 2026-09-30 |
| Value-Aware Stochastic KV Cache Eviction for Reasoning Models | ✓ Link | 72.29 | VaSE-AttnV (Qwen3-14B) | 2026-06-02 |
| Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements | ✓ Link | 71.7 | Agentic ESOpt + Trace2Skill (Qwen3.5-4B) | 2026-08-18 |
| Self-Organizing Agent Teams Learn to Reason Together | | 71.2 | Self-Organizing Agent Teams (SAT) | 2026-09-19 |
| Not Every Token Is Worth Distilling: Selective Supervision for Direct-OPD | ✓ Link | 70.9 | Qwen3-8B + S²D-OPD (Nemotron-1.5B → QuestA-1.5B) | 2026-09-26 |
| Is Next-Chunk Reasoning RL Really Better than SFT? Revisiting Training Strategies under no-CoT Data | | 70.42 | Mixed SFT + RLVR | 2026-08-24 |
| Surprising Success, Repeated Failure: Entropy-Guided Credit Assignment for Exploration in LLM Reasoning | ✓ Link | 70.3 | Olmo-3-7B-Think-DPO + EAPO | 2026-09-27 |
| DiffusionGemma Technical Report | ✓ Link | 69.1 | DiffusionGemma 26B A4B | 2026-07-31 |
| Where the Model Changes Its Mind: Hindsight-Divergence Localization for Efficient Reinforcement Learning with Verifiable Rewards | | 69.1 | HDL (Qwen3-8B) | 2026-09-29 |
| Not All Tokens Deserve Equal Credit: Counterfactual Sensitivity Credit Reallocation for Long-CoT Reasoning | | 67.2 | CSCR (Qwen3-4B) | 2026-07-30 |
| MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training | | 65.31 | Qwen3-30B-A3B + MOPD | 2026-06-29 |
| Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation | ✓ Link | 63.75 | Waypoint-OPD (Qwen3-14B) | 2026-10-15 |
| ExpRL: Exploratory RL for LLM Mid-Training | ✓ Link | 63.41 | ExpRL-Process (Ours) | 2026-06-15 |
| Calibrating Teacher--Student Discrepancy for On-Policy Distillation | | 60.2 | Cal-OPD (Qwen3-30B-A3B-Thinking-2507 -> Qwen3-4B) | 2026-09-18 |
| Can We Trust the Teacher? Decoupled Credit Direction-Magnitude for Self-Distillation | | 59.17 | DCSD (Qwen3-4B) | 2026-09-28 |
| Every Coin Has Two Sides: On the Dual Nature of Generalization in On-Policy Distillation of Large Language Models | | 50.6 | DS-R1-Distill-Qwen-7B + OPD Dynamic Sampling | 2026-08-17 |
| LFM2.5-8B-A1B: an Even Better on-Device Mixture-of-Experts | ✓ Link | 50.00 | LFM2.5-8B-A1B | 2026-05-28 |
| CoTranslator: Reverse to Forward Chain-of-Thought Translator for Reasoning Distillation | | 46.51 | CoTranslator (Qwen3-4B-Thinking-2507) | 2026-10-14 |
| Hybrid Reasoning Agents | ✓ Link | 45.8 | HybridAgent (14B) | 2026-10-13 |
| DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards | ✓ Link | 45.21 | DelTA (Qwen3-14B-Base) | 2026-05-20 |
| Selection Is Not Learning: Choosing Model Futures | | 41.88 | EvoHeir (Gemma-4-E4B-it) | 2026-10-15 |
| ARC: Fair Relative Advantage Comparison in Open-Ended Real-World Interaction | | 41.25 | DAPO + ARC (Qwen3-8B) | 2026-08-13 |
| Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models | | 41.0 | MELT-1.6B | 2026-05-08 |
| MiniCPM5-1B | ✓ Link | 40.42 | MiniCPM5-1B (Thinking) | 2026-05-25 |
| Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients | | 40.0 | ZPPO (Qwen3.5-2B) | 2026-06-16 |
| TEMPO: Scaling Test-time Training for Large Reasoning Models | ✓ Link | 38.8 | Qwen3-14B-Base (TEMPO) | 2026-04-21 |
| Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability | | 38.44 | Strict top-16 (Qwen3-235B-A22B-Instruct-2507 -> Granite-4.1-8B) | 2026-10-06 |
| Demystifying On-Policy Distillation: Roles, Pathologies, and Regulations | | 36.0 | Qwen3-1.7B (OPD+Clip) | 2026-07-15 |