BlogLLM Benchmark Scores as Published, Side by Side
COLLIE: Results as Published
Every result the catalog holds for COLLIE: 15 models from 7 developers, read from 4 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area Instruction FollowingMaintainer Shunyu Yao and co-authorsIntroduced 2023Data version 2026103001, as of 2026-10-30
COLLIE is a grammar-based framework for building text-generation tasks whose constraints can be composed. Constraints cover word, sentence, paragraph, and passage levels, including counts, positions, and words that must appear or must not appear. COLLIE-v1 has 2,080 instances across 13 constraint structures, with target values taken from text corpora. A parser marks each generation true or false against the constraint, and the main score is the average success rate.
COLLIE
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5OpenAI, released 2025-08-07 | 99% | OpenAI, Introducing GPT-5checked 2026-10-15 | Effort: high |
| 70.5% | OpenAI, Introducing GPT-5checked 2026-10-15 | Effort: none | |
| o3OpenAI, released 2025-04-16 | 98.4% | OpenAI, Introducing GPT-5checked 2026-10-15 | Effort: high |
| GPT-4oOpenAI, released 2024-05-13 | 61% | OpenAI, Introducing GPT-5checked 2026-10-15 | - |
| Claude Sonnet 4.6Anthropic, released 2026-02-17 | 67.7% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max |
| Claude Sonnet 4.5Anthropic, released 2025-09-29 | 90.5% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max |
| Qwen3.5-397B-A17BAlibaba, released 2026-02-16 | 88.9% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max |
| Qwen3-30B-A3B-Instruct-2507Alibaba, released 2025-07-30 | 28% | AI21 Labs, AI21 Jamba2 Mini model cardchecked 2026-10-15 | - |
| Kimi K2.5Moonshot AI, released 2026-01-27 | 87.8% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max |
| GLM-5Z.ai, released 2026-02-12 | 86.4% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max |
| Mistral Medium 3.5Mistral AI, released 2026-05-22 | 95.8% | Mistral AI, Mistral Medium 3.5 model cardchecked 2026-10-15 | Effort: max |
| Mistral Small 4Mistral AI, released 2026-03-16 | 62.9% | Mistral AI, Mistral Small 4 model cardchecked 2026-10-15 | Effort: high |
| Ministral 3 14B Instruct 2512Mistral AI, released 2025-12-02 | 26% | AI21 Labs, AI21 Jamba2 Mini model cardchecked 2026-10-15 | - |
| Magistral Medium 1.2Mistral AI, released 2025-09-17 | 61.3% | Mistral AI, Mistral Small 4 model cardchecked 2026-10-15 | - |
| Magistral Small 1.2Mistral AI, released 2025-09-17 | 60.3% | Mistral AI, Mistral Small 4 model cardchecked 2026-10-15 | - |
| Nemotron 3 Nano 30B-A3BNVIDIA, released 2025-12-15 | 30% | AI21 Labs, AI21 Jamba2 Mini model cardchecked 2026-10-15 | Effort: non-thinking |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.