| DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression | ✓ Link | 60.5 | DeepSeek-V4.1-Flash-1017 | 2026-09-17 |
| DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression | ✓ Link | 54.8 | DeepSeek-V4.1-Flash Max | 2026-09-17 |
| Atria Dawn: The Dawn of Agentic Superintelligence | ✓ Link | 53.8 | Atria Dawn Preview | 2026-09-14 |
| MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement | ✓ Link | 53.1 | MiMo-V2.6-Pro-RL | 2026-10-08 |
| GLM-5.3-Flash: Frontier Intelligence, Flash Cost | ✓ Link | 48.8 | GLM-5.3-Flash | 2026-08-26 |
| GLM-5.3: Frontier Coding with Emergent Cyber Capabilities | ✓ Link | 48.2 | GLM-5.3 | 2026-08-14 |
| REMORY: Learning Residual Memory for Context Compaction | ✓ Link | 45.3 | Qwen3.8-27B + REMORY | 2026-10-08 |
| Introducing Beam: Reflection's 501B open-weight model | | 37.0 | Beam | 2026-10-05 |
| CompoWorld: Compositional Environment Scaling for General Agents | ✓ Link | 32.33 | CompoWorld | 2026-09-27 |
| Hy4 preview | ✓ Link | 32.1 | Hy4 preview | 2026-08-28 |
| DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence | ✓ Link | 31.8 | DeepSeek-V4-Pro-0813 (Max) | 2026-04-26 |
| WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution | | 28.67 | WorkRover-35B-A3B | 2026-10-13 |
| LFM2.5-24B-A2B: Built for Fast On-Device Agents | | 27.69 | LFM2.5-24B-A2B | 2026-10-13 |
| Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work | ✓ Link | 27.6 | Occamy-1.0 | 2026-09-04 |
| Introducing K2 Horizon: Frontier Performance, Radically Open | ✓ Link | 25.3 | K2-Horizon-375B-A23B | 2026-09-03 |
| DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence | ✓ Link | 25.1 | DeepSeek-V4-Flash-0731 (Max) | 2026-04-26 |
| Interlock: From Local Correctness to Task Completion in Long-Horizon Agents | | 18.33 | Qwen3.6-27B (Interlock-RFT) | 2026-10-14 |
| Skill2Env: Capability-Oriented Environment Synthesis from Skills for General Agents | | 17.7 | Skill2Env (ours / 35B-A3B) | 2026-09-27 |
| Trajectory-Level Speculation with Semantic Verification for Tool-Using Agents | ✓ Link | 10.5 | SpecAgent | 2026-10-15 |