Finance Agent v2
Professional Work · 2026-05-19
Finance Agent v2's headline leaderboard evaluates agents on the 450-question held-out Test split of expert-reviewed financial-analysis tasks. Scores average partial credit across three runs per question. The full dataset also contains 27 public and 450 private-validation questions, which are not part of this reported measurement.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3.8 Flash | 61.4 |
| Muse Spark Open | 61.3 |
| Muse Spark 1.2 | 60.6 |
| Muse Spark 1.3 | 60.0 |
| Muse Wonder Full | 59.1 |
| Gemini 3.7 Flash | 59.0 |
| Gemini 3.5 Flash | 57.9 |
| Qwen3.8 Max (0902) | 57.9 |
| GLM-5.3 Flash | 57.9 |
| MiMo-V2.6-Pro | 57.3 |
| Muse Spark 1.1 | 57.2 |
| DeepSeek V4.1 Flash 1017 | 57.0 |
| Muse Wonder | 56.6 |
| Gemini 3.6 Flash | 56.3 |
| MiMo-V2.6-Flash | 56.3 |
| GLM-5.3 | 55.8 |
| Hy4 Preview | 55.1 |
| GPT-5.6 Luna | 55.0 |
| Mistral Large 4 Preview | 54.7 |
| GPT-5.6 Terra | 54.4 |
| Claude Haiku 5.5 | 54.1 |
| Opus 4.8 | 53.9 |
| Sonnet 5 | 53.9 |
| GPT-5.6 Sol | 53.8 |
| Grok 4.6 | 53.7 |
| DeepSeek V4.1 Flash | 53.5 |
| Kimi K3 | 53.1 |
| Grok 4.7 | 52.3 |
| GPT-6.1 Sol | 52.0 |
| Ember-1 | 51.9 |
| Opus 4.7 | 51.5 |
| Step 5 Preview | 50.7 |
| Qwen3.8-Max | 50.6 |
| DeepSeek V4 Pro 0813 | 50.4 |
| GPT-6 Luna | 49.9 |
| GLM-5.2 | 49.7 |
| DeepSeek-V4-Flash-0731 | 49.5 |
| GPT-6 Sol | 49.0 |
| Qwen3.8 27B | 48.6 |
| Grok 4.5 | 48.3 |