Structured Output Benchmark (SOB): Value Accuracy
Agents · 2026-04-28
Value Accuracy is the Structured Output Benchmark metric for exact leaf-value agreement with verified ground truth. Missing paths count as wrong, and results are schema-complexity-weighted across 5,000 text, 209 image, and 115 audio records.
Top models (higher is better)
| Model | Score |
|---|---|
| Interfaze Beta | 80.5 |
| Gemini 3.5 Flash | 80.2 |
| Sonnet 5 | 79.3 |
| Grok 4.3 | 78.4 |
| Gemini 3 Flash Preview | 77.3 |
| GPT-5.4 Mini | 75.1 |