AdvancedIF (rubric mean score)
Chat & Writing
Mean share of satisfied rubric items per response on AdvancedIF's 1,645 prompts. It grants partial credit on the same responses as the pass@1 headline, so it is kept as a diagnostic.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3.1 Pro Preview | 97.1 |
| Gemini 3.6 Flash | 96.6 |
| DeepSeek V4.1 Flash | 96.3 |
| Claude Opus 5 | 95.5 |
| Grok 4.6 | 95.5 |
| Claude Fable 5.1 | 95.4 |
| GPT-5.5 | 95.3 |
| GLM-5.3 | 95.0 |
| GPT-5.4 | 95.0 |
| Grok 4.5 | 95.0 |
| Qwen3.5 397B A17B | 94.9 |
| GPT-6.1 Sol | 94.8 |
| GPT-6 Sol | 94.7 |
| GPT-5.6 Sol Pro | 94.6 |
| GPT-6 Astra | 94.6 |
| GLM-5.2 | 94.4 |
| Muse Spark 1.1 | 94.4 |
| Grok 4.7 | 94.3 |
| Gemini 3.5 Flash | 94.1 |
| Claude Haiku 5.5 | 94.0 |
| Claude Sonnet 5.5 | 94.0 |
| Opus 4.8 | 93.5 |
| Claude Opus 5.5 | 93.5 |
| GPT-5.6 Terra | 93.5 |
| Inkling | 93.5 |
| Opus 4.7 | 93.2 |
| GPT-6 Luna | 93.2 |
| Kimi K2.7 Code | 93.1 |
| GPT-5.6 Luna | 93.0 |
| Sonnet 5 | 92.9 |
| Sonnet 4.6 | 92.4 |
| Kimi K2 Thinking | 92.4 |
| gpt-oss-120b | 90.9 |
| MiniMax M3 | 88.4 |
| DeepSeek-V3.2 | 87.8 |
| MiniMax M2.7 | 84.2 |
| Kimi K3 | 80.7 |
| Nemotron 3 Ultra 550B A55B (NVFP4) | 80.3 |
| Claude Fable 5 | 79.0 |