Benchmark
IFBench
What is IFBench?
Instruction Following Benchmark evaluating model's ability to follow complex instructions. Benchmark Radar tracks 34 reported scores for IFBench, cited by 1…
- Released
- 2025-07-03
- Modality
- text
- Evaluates
- instruction following, general
- Openness
- open
- Importer
- LLM Stats
- Reported scores
- 34
Source provenance
- Original evidence https://api.zeroeval.com/leaderboard/benchmarks/ifbench?top_n=500
IFBench paper, code and dataset
IFBench results and reported scores
LLM Stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| Command A+ | Cohere | 0.74 | 2026-05-20 | evidence |
| GPT OSS 120B High | OpenAI | 0.695 | 2025-08-05 | evidence |
| Hermes 3 70B | Nous Research | 0.8121 | 2024-08-15 | evidence |
| Inkling-Small | Thinking Machines Lab | 0.822 | 2026-07-30 | evidence |
| K-EXAONE-236B-A23B | LG AI Research | 0.673 | 2025-12-31 | evidence |
| LFM2.5-2.6B | Liquid AI | 0.5917 | 2026-08-04 | evidence |
| LFM2.5-VL-3B | Liquid AI | 0.258 | 2026-08-12 | evidence |
| MAI-Code-1-Flash | Microsoft | 0.75 | 2026-06-02 | evidence |
| MAI-Thinking-1 | Microsoft | 0.69 | 2026-06-02 | evidence |
| Mercury 2 | Inception | 0.71 | 2026-02-24 | evidence |
| MiniMax M2.1 | MiniMax | 0.7 | 2025-12-23 | evidence |
| Mistral Medium 3.5 | Mistral | 0.69 | 2026-04-29 | evidence |
| Mistral Small 4 | Mistral | 0.48 | 2026-03-16 | evidence |
| Muse Glimmer-30B | Meta | 0.77 | 2026-08-10 | evidence |
| Nemotron 3 Super (120B A12B) | NVIDIA | 0.7256 | 2026-03-11 | evidence |
| Nemotron 3 Ultra (550B A55B) | NVIDIA | 0.817 | 2026-06-04 | evidence |
| Nemotron 3.5 Lightning (30B A3B) | NVIDIA | 0.7188 | 2026-08-11 | evidence |
| Nova 2 Lite | Amazon | 0.708 | 2025-12-02 | evidence |
| Nova 2 Omni | Amazon | 0.687 | 2025-12-02 | evidence |
| Nova 2 Pro | Amazon | 0.802 | 2025-12-02 | evidence |
| Nova 2 Sonic | Amazon | 0.375 | 2025-12-02 | evidence |
| Qwen3.5-0.8B | Qwen | 0.21 | 2026-03-02 | evidence |
| Qwen3.5-122B-A10B | Qwen | 0.761 | 2026-02-24 | evidence |
| Qwen3.5-27B | Qwen | 0.765 | 2026-02-24 | evidence |
| Qwen3.5-2B | Qwen | 0.413 | 2026-03-02 | evidence |
| Qwen3.5-35B-A3B | Qwen | 0.702 | 2026-02-24 | evidence |
| Qwen3.5-397B-A17B | Qwen | 0.765 | 2026-02-16 | evidence |
| Qwen3.5-4B | Qwen | 0.592 | 2026-03-02 | evidence |
| Qwen3.5-9B | Qwen | 0.645 | 2026-03-02 | evidence |
| Qwen3.6 Plus | Qwen | 0.742 | 2026-04-02 | evidence |
| Qwen3.7 Max | Qwen | 0.791 | 2026-05-19 | evidence |
| Qwen3.7-Plus | Qwen | 0.791 | 2026-05-31 | evidence |
| Qwen3.8 Max | Qwen | 0.828 | 2026-08-02 | evidence |
| Qwen3.8-27B | Qwen | 0.795 | 2026-08-14 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.