JSONSchemaBench: leaderboard
Structured-output benchmark measuring schema compliance and JSON validity for language models across easy, medium, and hard schema-constrained generation datasets.
Metric: Schema Compliance (percentage). Source: github.com. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 45 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 96.9 | #343 |
| 2 | GPT-4o Mini | 95.8 | #599 |
| 3 | Qwen 2.5 72B Instruct | 95.5 | #454 |
| 4 | Qwen 2.5 32B Instruct | 94.3 | #498 |
| 5 | Qwen 2.5 Coder 7B Instruct | 94.3 | #951 |
| 6 | Qwen 2.5 14B Instruct | 92.1 | #647 |
| 7 | Qwen 2.5 Coder 3B Instruct | 91.2 | #1026 |
| 8 | Llama 3.1 8B Instruct | 91.1 | #1035 |
| 9 | Llama 3.2 3B Instruct | 88.3 | #1352 |
| 10 | Qwen 2.5 3B Instruct | 83.8 | #1162 |
| 11 | Phi-4 Mini Instruct | 81.3 | #1245 |
| 12 | Phi-3.5-mini-instruct | 77.2 | #1167 |
| 13 | Gemma 3 1B (IT) | 69.9 | #1504 |
| 14 | Llama 3.2 1B Instruct | 65.6 | #1539 |
| 15 | Qwen 2.5 7B Instruct | 54.8 | #862 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=jsonschemabench · How It Works · Data refreshed daily, snapshot 2026-10-20.