GRIPS-hard: leaderboard
The hard subset of GRIPS, restricted to the German reasoning, idiom and wordplay items that discriminate most between models; scores drop sharply versus the full set.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around January 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 87.3 | #31 (GPT-5.5) |
| 2 | GPT-5.5 (High) | 85.3 | #31 (GPT-5.5) |
| 3 | GPT-5.5 | 80.7 | #31 |
| 4 | Claude Opus 4.8 (High) | 76 | #37 (Claude Opus 4.8) |
| 5 | Gemini 3.5 Flash (High) | 74 | #60 (Gemini 3.5 Flash) |
| 6 | GPT-5.5 (Low) | 70.7 | #31 (GPT-5.5) |
| 7 | DeepSeek V4 Pro (Reasoning) | 67.3 | #102 (DeepSeek V4 Pro) |
| 8 | Gemini 3.5 Flash (Low) | 65.3 | #60 (Gemini 3.5 Flash) |
| 9 | Claude Sonnet 4.6 (High) | 60.4 | #90 (Claude Sonnet 4.6) |
| 10 | DeepSeek V4 Flash (Reasoning) | 49.3 | #137 (DeepSeek V4 Flash) |
| 11 | GPT-5.4 Mini (High) | 42 | #212 (GPT-5.4 Mini) |
| 12 | Mistral Medium 3.5 (High) | 26.2 | #353 (Mistral Medium 3.5) |
| 13 | GPT-5.4 Mini | 17.3 | #212 |
| 14 | GPT-5.4 Mini (Low) | 16 | #212 (GPT-5.4 Mini) |
| 15 | Claude Haiku 4.5 | 14 | #280 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=grips-hard · How It Works · Data refreshed daily, snapshot 2026-10-20.