TaxEval v2 - Correctness
Professional Work · 2025-04-18
The Correctness split of TaxEval v2. This child benchmark separates a source-reported subtask or subtrack from the parent aggregate so scores at different grains do not share one benchmark_slug.
Top models (higher is better)
| Model | Score |
|---|---|
| Grok 4 Fast | 68.2 |
| Muse Spark 1.1 | 66.9 |
| GPT-5.6 Luna | 66.6 |
| GPT-5.2 | 66.3 |
| Claude Fable 5 | 65.8 |
| Sonnet 4.6 | 65.8 |
| GPT-5.6 Terra | 65.8 |
| Muse Spark | 65.6 |
| GPT-5.5 | 65.5 |
| Kimi K3 | 65.3 |
| GPT-5.6 Sol | 65.0 |
| GPT-5.4 | 64.9 |
| GPT-4o (2024-11-20) | 64.8 |
| GPT-5.1 | 64.8 |
| Opus 4.6 | 64.8 |
| GPT-5 Mini | 64.8 |
| Opus 4.8 | 64.5 |
| O1 | 64.4 |
| O4 Mini | 64.4 |
| Qwen3.6 Plus (2026-04-02) | 64.2 |
| Sonnet 5 | 64.0 |
| o3 | 64.0 |
| Opus 4.5 | 63.9 |
| GPT-4.1 | 63.9 |
| Grok 3 | 63.9 |
| Opus 4.7 | 63.8 |
| Gemini 3.6 Flash | 63.5 |
| Inkling | 63.5 |
| Qwen3.7-Max | 63.5 |
| Grok 4.20 | 63.4 |
| Claude 3.7 Sonnet | 63.3 |
| MiMo-V2.5-Pro | 63.1 |
| GPT-5.4 Mini | 63.0 |
| Opus 4.1 | 63.0 |
| GPT-5 | 62.8 |
| Kimi K2.6 | 62.7 |
| Grok 4.5 | 62.6 |
| Qwen3 Max Preview | 62.4 |
| Gemini 3 Flash Preview | 62.3 |
| GLM-5.2 | 62.1 |