Share your thoughts, 1 month free Claude Pro on usSee more

Structured JSON Generation on JSON Generation dataset

99.8Structural Accuracy

Qwen3-4B + Outlines

Updated 5mo ago

Evaluation Results

Method	Links
Qwen3-4B + Outlines 2025.11		99.8	100	99.5	99.8	79.5	15.2	2.1
RL-Struct 2025.11		89.7	92.1	85.3	89.7	84.5	90.6	1.5
Qwen3-4B + PPO 2025.11		89.1	91.5	84.8	89	84.2	90.4	1.8
Qwen3-4B + DPO 2025.11		82.5	88.4	83	81.5	82	90.2	6.5
LLaMA-3-8B 2025.11		78.2	85.4	81.2	78.2	86	40.1	9.3
Phi-3-mini 2025.11		74.1	84.8	79.5	74.1	81.5	85.4	8.1
Qwen3-4B 2025.11		65.4	72.1	68.9	68.5	80	90.8	14.1
Mistral-7B 2025.11		52.3	83.5	76.2	60.5	85	45.2	12.5
GPT-3.5 2025.11		45.5	82.1	75	55.2	88	-	15.2