APEX-Accounting 1.1
Finance & Accounting
Mercor and Ramp's 100-task APEX-Accounting cohort, which Mercor's benchmark map calls APEX-Accounting 1.1: held-out month-end-close tasks across 10 accounting worlds, scored by an LM judge against 1,475 expert-authored criteria. Mean Score averages the fraction of criteria passed per task using the Loop agent; per-model run counts are not published and the pass curves stop at k=4. Kept separate from the 160-task, 2,186-criterion APEX-Accounting cohort.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 71.5 |
| Gemini 4 Carbon | 70.7 |
| Muse Wonder | 67.1 |
| Claude Fable 5.1 | 66.7 |
| Claude Opus 5.5 | 65.5 |
| Muse Spark 1.3 | 64.9 |
| GPT-6 Astra | 64.7 |
| Claude Sonnet 5.5 | 64.5 |
| Claude Opus 5 | 64.3 |
| GPT-6.1 Sol | 64.3 |
| GPT-6.1 Terra | 63.7 |
| Kimi K3.1 Pro | 62.4 |
| Claude Haiku 5.5 | 62.2 |
| GPT-5.6 Sol Pro | 62.2 |
| Claude Fable 5 | 61.0 |
| Grok 4.6 | 60.3 |
| MiniMax M3.1 Flash | 60.0 |
| GPT-5.6 Terra | 59.8 |
| GPT-6 Sol | 59.7 |
| GLM-5.3 | 59.5 |
| DeepSeek V4.1 Flash 1017 | 59.3 |
| Gemini 3.8 Flash | 58.8 |
| Muse Spark Open | 58.0 |
| Grok 4.5 | 57.4 |
| Gemini 3.7 Flash | 57.2 |
| Kimi K3 | 56.7 |
| Opus 4.7 | 56.5 |
| Mistral Large 4 Preview | 56.5 |
| GPT-5.5 | 56.3 |
| Sonnet 5 | 55.7 |
| Grok 4.7 | 55.3 |
| GPT-5.6 Luna | 54.3 |
| GPT-6 Luna | 53.9 |
| GLM-5.3 Flash | 53.7 |
| DeepSeek V4 Pro 0813 | 53.4 |
| Gemini 3.6 Flash | 52.5 |
| Mercury 3 | 50.8 |
| DeepSeek V4.1 Flash | 50.1 |
| MiniMax M3 | 41.8 |
| Gemini 3.5 Flash | 31.4 |