APEX-Accounting Pass@1
Professional Work · 2026-07-31
Strict task success on APEX-Accounting's 160 month-end-close tasks: every criterion must pass in one attempt. Kept separate from the partial-credit Mean Score.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 14.8 |
| GPT-6 Astra | 11.9 |
| Claude Fable 5.1 | 11.7 |
| Muse Spark 1.3 | 10.8 |
| Claude Opus 5 | 10.5 |
| Claude Fable 5 | 9.8 |
| GPT-5.6 Sol Pro | 9.6 |
| GLM-5.3 | 9.2 |
| Muse Spark 1.2 | 8.9 |
| Gemini 3.8 Flash | 8.7 |
| GPT-5.6 Sol | 8.4 |
| DeepSeek V4.1 Flash | 7.8 |
| Muse Spark 1.1 | 7.7 |
| GPT-5.5 | 6.2 |
| Kimi K3 | 6.2 |
| GPT-5.6 Terra | 6.0 |
| GPT-5.4 | 5.3 |
| Grok 4.5 | 4.5 |
| Opus 4.6 | 4.4 |
| GLM-5.2 | 4.4 |
| GPT-5.6 Luna | 4.4 |
| Opus 4.8 | 4.2 |
| Grok 4.7 | 4.2 |
| Gemini 3.1 Pro Preview | 3.1 |
| Kimi K2.7 Code | 2.9 |
| MiniMax M3 | 2.2 |
| GLM-5.3 Flash | 1.7 |
| GLM-5.1 | 1.2 |
| DeepSeek-V3.2 | 1.1 |
| Inkling | 0.8 |
| Nemotron 3 Ultra 550B A55B (NVFP4) | 0.3 |
| gpt-oss-120b | 0.0 |
| Qwen3.5 397B A17B | 0.0 |