Simple Retrieval on Ruler S-NIAH
100AccuracyGPT-5
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5Architecture=Base LLM2026.03 | 100 | |
| RLMBackbone=GPT-5, Recursion Depth=12026.03 | 100 | |
| DeepSeek v3.2Architecture=Base LLM2026.03 | 100 | |
| Kimi K2Architecture=Base LLM2026.03 | 100 | |
| RLMBackbone=Kimi K2, Recursion Depth=12026.03 | 90 | |
| RLMBackbone=Kimi K2, Recursion Depth=22026.03 | 90 | |
| RLMBackbone=DeepSeek v3.2, Recursion Depth=12026.03 | 85 | |
| RLMBackbone=DeepSeek v3.2, Recursion Depth=22026.03 | 70 |