DualEntry Accounting AI Benchmark Consistency
Finance & Accounting
Share of DualBench100's 101 accounting cases a model passes in all three runs. Kept separate from accuracy, which needs two of three.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 86.1 |
| Claude Fable 5.1 | 80.2 |
| Claude Opus 5.5 | 78.2 |
| Kimi K3 | 78.2 |
| Muse Wonder | 78.2 |
| Gemini 3.8 Flash | 77.2 |
| Grok 4.5 | 77.2 |
| GLM-5.3 Flash | 76.2 |
| GLM-5.3 | 75.2 |
| Claude Sonnet 5.5 | 74.3 |
| Grok 4.7 | 74.3 |
| DeepSeek V4.1 Flash | 73.3 |
| Kimi K3.1 Pro | 73.3 |
| GPT-6 Astra | 71.3 |
| Claude Haiku 5.5 | 69.3 |
| GPT-6 Sol | 69.3 |
| Mistral Large 4 Preview | 69.3 |
| GPT-6.1 Sol | 68.3 |
| GPT-6.1 Terra | 65.3 |
| GPT-6 Luna | 49.5 |