GRIPS: leaderboard
The model answers German-language puzzle questions covering anagrams, hidden words, idioms, compound words and multi-step logic riddles. The published set holds 1,861 items across 96 mechanics, but the leaderboard is run on a held-out private set of unreleased items. An LLM judge compares each answer with the reference, accepting equivalent wording, and the score is the share judged correct; higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around January 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 96.4 | #31 (GPT-5.5) |
| 2 | GPT-5.5 (High) | 96.1 | #31 (GPT-5.5) |
| 3 | GPT-5.5 | 95.9 | #31 |
| 4 | Gemini 3.5 Flash (High) | 94.8 | #60 (Gemini 3.5 Flash) |
| 5 | Claude Opus 4.8 (High) | 93.7 | #37 (Claude Opus 4.8) |
| 6 | GPT-5.5 (Low) | 93.6 | #31 (GPT-5.5) |
| 7 | DeepSeek V4 Pro (Reasoning) | 92.9 | #102 (DeepSeek V4 Pro) |
| 8 | Claude Sonnet 4.6 (High) | 92.6 | #90 (Claude Sonnet 4.6) |
| 9 | Gemini 3.5 Flash (Low) | 91.6 | #60 (Gemini 3.5 Flash) |
| 10 | DeepSeek V4 Flash (Reasoning) | 88.2 | #137 (DeepSeek V4 Flash) |
| 11 | GPT-5.4 Mini (High) | 85.9 | #212 (GPT-5.4 Mini) |
| 12 | Mistral Medium 3.5 (High) | 79.2 | #353 (Mistral Medium 3.5) |
| 13 | GPT-5.4 Mini (Low) | 73.8 | #212 (GPT-5.4 Mini) |
| 14 | Mistral Medium 3.5 | 59 | #353 |
| 15 | Claude Haiku 4.5 | 52.7 | #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 · How It Works · Data refreshed daily, snapshot 2026-10-20.