JSONSchemaBench
Models or constrained decoders generate outputs under real JSON schemas. The evaluation tests schema coverage and valid generation.
- Made by
- Saibo Geng and colleagues
- Released
- January 2025
- Kind
- Following precise instructions
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
-
Qwen2.5-72B #166 on the capability index
- Last model added
- March 2025
- Scores refreshed
- November 20th, 2026
GPT-4o leads with 91.4% among models released by March 2025 GPT-4o leads with 91.4% among models released by March 2025
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in March 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 91.4 GPT-4o
- 85.3 Qwen/Qwen2.5-32B-Instruct
- 83.5 GPT-4o mini
- 82.1
Qwen2.5-72B
- 78.2 Qwen/Qwen2.5-Coder-7B-Instruct
- 74.2 Qwen/Qwen2.5-14B-Instruct
- 69.8 meta-llama/Llama-3.1-8B-Instruct
- 69.0 Qwen2.5-Coder-3B-Instruct
- 63.2 Qwen/Qwen2.5-3B-Instruct
- 61.9 meta-llama/Llama-3.2-3B-Instruct
- 57.7 microsoft/Phi-4-mini-instruct
- 56.1 microsoft/Phi-3.5-mini-instruct
- 43.1
Gemma 3 1B
- 39.7 meta-llama/Llama-3.2-1B-Instruct
- 31.1 Qwen/Qwen2.5-7B-Instruct
- 1 GPT-4o Mar 2025 91.4%
- 2 Qwen/Qwen2.5-32B-Instruct 85.3%
- 3 GPT-4o mini 83.5%
- 4
Qwen2.5-72B 82.1%
- 5 Qwen/Qwen2.5-Coder-7B-Instruct 78.2%
- 6 Qwen/Qwen2.5-14B-Instruct 74.2%
- 7 meta-llama/Llama-3.1-8B-Instruct 69.8%
- 8 Qwen2.5-Coder-3B-Instruct 69%
- 9 Qwen/Qwen2.5-3B-Instruct 63.2%
- 10 meta-llama/Llama-3.2-3B-Instruct 61.9%
- 11 microsoft/Phi-4-mini-instruct 57.7%
- 12 microsoft/Phi-3.5-mini-instruct 56.1%
- 13
Gemma 3 1B 43.1%
- 14 meta-llama/Llama-3.2-1B-Instruct 39.7%
- 15 Qwen/Qwen2.5-7B-Instruct 31.1%
Source: Official JSONSchemaBench leaderboard
- 1 GPT-4o Mar 2025 91.4%
- 2 GPT-4o mini 83.5%
- 3
Qwen2.5-72B 82.1%
- 4 Qwen2.5-Coder-3B-Instruct 69%
- 5
Gemma 3 1B 43.1%
Source: Official JSONSchemaBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o Mar 2025 | 91.4% | Mar 2025 | Official leaderboard | |
| 2 | Qwen/Qwen2.5-32B-Instruct | 85.3% | Official leaderboard | ||
| 3 | GPT-4o mini | 83.5% | Jul 2024 | Official leaderboard | |
| 4 | | 82.1% | Sep 2024 | Official leaderboard | |
| 5 | Qwen/Qwen2.5-Coder-7B-Instruct | 78.2% | Official leaderboard | ||
| 6 | Qwen/Qwen2.5-14B-Instruct | 74.2% | Official leaderboard | ||
| 7 | meta-llama/Llama-3.1-8B-Instruct | 69.8% | Official leaderboard | ||
| 8 | Qwen2.5-Coder-3B-Instruct | 69% | Nov 2024 | Official leaderboard | |
| 9 | Qwen/Qwen2.5-3B-Instruct | 63.2% | Official leaderboard | ||
| 10 | meta-llama/Llama-3.2-3B-Instruct | 61.9% | Official leaderboard | ||
| 11 | microsoft/Phi-4-mini-instruct | 57.7% | Official leaderboard | ||
| 12 | microsoft/Phi-3.5-mini-instruct | 56.1% | Official leaderboard | ||
| 13 | | 43.1% | Mar 2025 | Official leaderboard | |
| 14 | meta-llama/Llama-3.2-1B-Instruct | 39.7% | Official leaderboard | ||
| 15 | Qwen/Qwen2.5-7B-Instruct | 31.1% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4o Mar 2025 | 91.4% | Mar 2025 | Official leaderboard | |
| 2 | GPT-4o mini | 83.5% | Jul 2024 | Official leaderboard | |
| 3 | | 82.1% | Sep 2024 | Official leaderboard | |
| 4 | Qwen2.5-Coder-3B-Instruct | 69% | Nov 2024 | Official leaderboard | |
| 5 | | 43.1% | Mar 2025 | Official leaderboard |
Rankings by part
We read 3 parts from the benchmark’s own leaderboard: Easy, Medium and Hard. Each ranks models on one part of the test. GPT-4o leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.
Easy
5 models we track have a score here.
- 1 GPT-4o Mar 2025 96.9%
- 2 GPT-4o mini 95.8%
- 3
Qwen2.5-72B 95.5%
- 4 Qwen2.5-Coder-3B-Instruct 91.2%
- 5
Gemma 3 1B 69.9%
Medium
5 models we track have a score here.
- 1 GPT-4o Mar 2025 89.6%
- 2 GPT-4o mini 86.2%
- 3
Qwen2.5-72B 84%
- 4 Qwen2.5-Coder-3B-Instruct 73.2%
- 5
Gemma 3 1B 37.1%
Hard
5 models we track have a score here.
- 1 GPT-4o Mar 2025 87.8%
- 2 GPT-4o mini 68.5%
- 3
Qwen2.5-72B 66.7%
- 4 Qwen2.5-Coder-3B-Instruct 42.5%
- 5
Gemma 3 1B 22.3%
2 models have held number one on the official leaderboard since its release in January 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official JSONSchemaBench leaderboard
GPT-4o on the official leaderboard passed 85% in March 2025
That was 2 months after its release. By the same age, 11% of the benchmarks from 2025 had passed 85%. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official JSONSchemaBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Instructions and chat
- IFEval In use Models answer about 500 prompts containing verifiable instructions such as word limits or required keywords. Code checks each instruction for strict and loose accuracy.
Gemma 4 31B IT 98.9%
- IFBench In use 58 new out-of-domain verifiable output constraints. Scored by programmatic constraint checks (strict or loose). Flint 89.2%
- MultiChallenge In use Multi-turn conversations testing instruction retention, inference memory, versioned editing and self-coherence. Graded by LLM judge with rubrics. GPT-6.1 Sol 82%
- Multi-IF In use Models follow verifiable instructions over three-turn conversations in eight languages, extending IFEval. Scored by instruction accuracy at each turn. LFM3-1.7B 89.1%
- AdvancedIF In use Models respond to over 1,600 prompts with complex, multi-turn, and system-level instructions. A judge model checks each response against expert-written rubrics.
MAI-Thinking-1 85%
- MM-IFEval In use Multimodal models follow image-grounded instructions with exact output constraints. Checks mix rule-based verification and a judge model. GPT-4o 64.6%
Data audit
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the official leaderboard (15 entries).
- Pass Models scored 5 models we track, plus 10 other entries on its leaderboard.
- Note Newest score The newest came 20 months ago, in March 2025, for GPT-4o. That’s expected for a saturated test.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Qwen2.5-72B (#166).
- Note Headroom Saturated. The best is 91.4%, past 85% since March 2025.
Leaderboard import 3 pass
- Pass Importer We read the board on docs.google.com: 5 rows for models we track and 10 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.