Horangi 4 - IFEval-Ko: leaderboard
IFEval-Ko, a Korean adaptation of IFEval with automatically verifiable instructions, as run by the Horangi Korean LLM Leaderboard 4 (W&B Korea). Score is the mean of strict and loose prompt- and instruction-level accuracy (%), higher is better.
Metric: Instruction-following accuracy (%). Source: horangi.ai. Saturation forecast: Estimated already saturated. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (2026-03-05) (xHigh) | 95 | #63 (GPT-5.4 (2026-03-05)) |
| 2 | GPT-5.1 (2025-11-13) | 93.75 | #107 |
| 3 | GPT-5.5 (xHigh) | 93.33 | #33 (GPT-5.5) |
| 4 | GPT-5 | 92.92 | #105 |
| 5 | GLM-5-Turbo | 92.92 | #121 |
| 6 | Grok 4.20 | 92.5 | #179 |
| 7 | Grok 4.5 | 92.5 | #54 |
| 8 | GPT-5.2 (2025-12-11) (xHigh) | 92.5 | |
| 9 | GPT-5.4 Mini (2026-03-17) (xHigh) | 92.5 | #180 (GPT-5.4 Mini (2026-03-17)) |
| 10 | GPT-5.4 Nano (2026-03-17) (xHigh) | 92.5 | #285 (GPT-5.4 Nano (2026-03-17)) |
| 11 | GPT-5.6 Sol (Max) | 91.67 | #22 (GPT-5.6 Sol) |
| 12 | Gemini 3.5 Flash | 91.25 | #66 |
| 13 | Gemini 3.1 Pro (Preview) (High) | 91.25 | #64 (Gemini 3.1 Pro (Preview)) |
| 14 | GPT-5.6 Terra (Max) | 91.25 | #40 (GPT-5.6 Terra) |
| 15 | Gemma 4 31B (IT) (Thinking) | 91.25 | #178 (Gemma 4 31B (IT)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=horangi-4-ifeval-ko · How It Works · Data refreshed daily, snapshot 2026-11-05.