Finance Agent v2 - All Pass
Professional Work · 2026-05-19
Finance Agent v2 strict All-Pass grading metric over all 450 held-out test questions. Every required answer component must pass; published results are means over three model runs.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 55.8 |
| Muse Spark 1.2 | 50.9 |
| Muse Spark Open | 50.6 |
| Gemini 3.8 Flash | 49.7 |
| Muse Spark 1.3 | 49.5 |
| Muse Wonder Full | 49.0 |
| Claude Fable 5.1 | 47.8 |
| Gemini 3.7 Flash | 47.5 |
| Qwen3.8 Max (0902) | 47.2 |
| GLM-5.3 Flash | 46.3 |
| Muse Wonder | 45.9 |
| MiMo-V2.6-Pro | 45.8 |
| Gemini 3.5 Flash | 45.7 |
| DeepSeek V4.1 Flash 1017 | 45.2 |
| Muse Spark 1.1 | 44.9 |
| GLM-5.3 | 44.9 |
| Gemini 3.6 Flash | 44.6 |
| MiMo-V2.6-Flash | 44.4 |
| Hy4 Preview | 44.0 |
| Mistral Large 4 Preview | 43.8 |
| Opus 4.8 | 43.5 |
| GPT-5.6 Luna | 43.3 |
| Grok 4.6 | 42.9 |
| Claude Haiku 5.5 | 42.9 |
| GPT-5.6 Terra | 42.8 |
| Grok 4.7 | 42.1 |
| Sonnet 5 | 42.1 |
| Kimi K3 | 41.8 |
| GPT-5.6 Sol | 41.4 |
| DeepSeek V4.1 Flash | 41.1 |
| GPT-6 Astra | 40.3 |
| GPT-6.1 Sol | 39.9 |
| GPT-5.5 | 39.6 |
| Ember-1 | 39.3 |
| Step 5 Preview | 38.8 |
| Sonnet 4.6 | 38.8 |
| Opus 4.7 | 38.6 |
| DeepSeek V4 Pro 0813 | 38.5 |
| GPT-6 Luna | 38.4 |
| Qwen3.8-Max | 38.2 |