AA-Briefcase v1.1 Elo
Agents
AA-Briefcase v1.1 evaluates 91 tasks across four multi-week professional projects. Its Elo combines rubric completion, analytical quality, and presentation quality. Tasks are run independently, without carrying over a model's earlier submissions. This protocol version is recorded separately from the original AA-Briefcase.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 1822 |
| Claude Fable 5.1 | 1678 |
| Claude Opus 5 | 1673 |
| Grok 4.7 | 1657 |
| Qwen3.8 Max (0902) | 1640 |
| Muse Spark 1.3 | 1597 |
| Qwen3.8-Flash-Next | 1597 |
| GPT-6 Astra | 1569 |
| Grok 4.6 | 1555 |
| Claude Fable 5 | 1543 |
| GLM-5.3 | 1525 |
| MiMo-V2.6-Pro | 1522 |
| Kimi K3 | 1510 |
| GPT-5.6 Sol | 1487 |
| GPT-6 Sol | 1483 |
| GLM-5.3 Flash | 1459 |
| Qwen3.8 2.4T A95B | 1445 |
| DeepSeek V4 Flash Vision Exp | 1437 |
| Step 5 Preview | 1432 |
| DeepSeek V4.1 Flash | 1432 |
| Qwen3.8 27B | 1403 |
| Qwen3.8-Max | 1390 |
| Sonnet 5 | 1363 |
| GPT-5.6 Luna | 1345 |
| Muse Spark 1.2 | 1340 |
| GPT-5.6 Terra | 1339 |
| Opus 4.8 | 1322 |
| K2 Horizon 375B A23B | 1303 |
| GPT-6 Luna | 1299 |
| Grok 4.5 | 1286 |
| DeepSeek V4 Pro 0813 | 1261 |
| DeepSeek-V4-Flash-0731 | 1260 |
| Opus 4.7 | 1255 |
| GLM-5.2 | 1233 |
| Gemini 3.8 Flash | 1202 |
| Agnes 2.5 Pro Alpha | 1197 |
| K2 Horizon MoVA 36B A4B | 1170 |
| GPT-5.5 | 1139 |
| Quasar 438B | 1122 |
| Gemini 3.7 Flash | 1108 |