Nejumi 4 - ALT - Controllability: leaderboard
Two instruction-following checks averaged. Jaster Control asks whether answers on jaster tasks use the required output format, such as digits only, a letter choice, a 0 or 1, or an entailment label, and is taken as the mean of the 0-shot and 2-shot runs; M-IFEval adds Japanese prompts whose constraints a program can verify. Reported in percent, higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Estimated already saturated. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.5 | 94.73 | #48 |
| 2 | GPT-5.6 Sol | 93.85 | #19 |
| 3 | Inkling | 93.54 | #126 |
| 4 | GPT-5.6 Terra | 93.44 | #41 |
| 5 | Qwen 3.6 Max Preview | 93.37 | #85 |
| 6 | DeepSeek V4.1 Flash | 93.18 | #36 |
| 7 | GPT-5.5 (2026-04-23) | 93.16 | #22 |
| 8 | Muse Glimmer 30B | 93.09 | #178 |
| 9 | GPT-5.4 (2026-03-05) | 92.98 | #59 |
| 10 | Gemini 3.1 Pro (Preview) | 92.74 | #58 |
| 11 | Gemini 3 Flash (Preview) | 92.74 | #82 |
| 12 | Hy3 | 92.37 | #93 |
| 13 | GPT-5.2 | 92.35 | #110 |
| 14 | Gemini 3.6 Flash | 92.32 | #51 |
| 15 | Inkling Small | 92.32 | #138 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=nejumi-4-alt-controllability · How It Works · Data refreshed daily, snapshot 2026-10-20.