language benchmark
COLLIE
COLLIE is a grammar-based framework for systematic construction of constrained text generation tasks. It allows specification of rich, compositional constraints across diverse generation levels and modeling challenges including language understanding, logical reasoning, and semantic planning. The COLLIE-v1 dataset contains 2,080 instances across 13 constraint structures.
Updated Aug 11, 2026
Models10
Model coverage10
MetricScore
EvidenceB
COLLIE Ranking
Higher score ranks better on this benchmark.
| 01 | OP | 99.0% | 100.0% | 10 | C | |
| 02 | OP | 98.7% | 88.9% | 10 | C | |
| 03 | OP | 98.4% | 77.8% | 10 | C | |
| 04 | MA | 95.8% | 66.7% | 10 | C | |
| 05 | OP | 72.3% | 55.6% | 10 | C | |
| 06 | OP | 65.8% | 44.4% | 10 | C | |
| 07 | MA | 62.9% | 33.3% | 10 | C | |
| 08 | OP | 61.0% | 22.2% | 10 | C | |
| 09 | OP | 54.6% | 11.1% | 10 | C | |
| 10 | OP | 42.5% | 0.0% | 10 | C |
COLLIE Score Distribution
A closer view of the leading scores on this benchmark.
COLLIE Highlights
The leading models and scores on this benchmark.
What is COLLIE?
What COLLIE measures and how its scores work.
COLLIE is a grammar-based framework for systematic construction of constrained text generation tasks. It allows specification of rich, compositional constraints across diverse generation levels and modeling challenges including language understanding, logical reasoning, and semantic planning. The COLLIE-v1 dataset contains 2,080 instances across 13 constraint structures.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- COLLIE
- Modality
- text
- Primary category
- language
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- collie|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about COLLIE.
Which model scores highest on COLLIE?
GPT-5 is currently ranked first with 99.0%.
What does COLLIE measure?
COLLIE is a grammar-based framework for systematic construction of constrained text generation tasks. It allows specification of rich, compositional constraints across diverse generation levels and modeling challenges including language understanding, logical reasoning, and semantic planning. The COLLIE-v1 dataset contains 2,080 instances across 13 constraint structures.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
10 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.