IFHierBench - Instruction-Level: leaderboard

Metric: Instruction-level strict accuracy (%; share of individual constraints satisfied, a failed outer container marking its descendants as failed; prompts whose output constraints are nested in a tree (artifact, sections, fields) of depth 0 to 3, checked by Python verifiers as in IFEval; 10 structure-level and 25 content-level constraint types). Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked.

Top models

#ModelScore
1GPT-5.5 (High)65.3
2Claude Opus 4.6 (High)61.9
3DeepSeek R137.7
4Kimi K2.533
5Gemini 3 Flash (Preview)24.1

Interactive version: theaggregate.ai/benchmark?slug=ifhierbench-instruction-level · How It Works · Data refreshed daily, snapshot 2026-09-29.