SloBench - DIALECT-COPA (Standard Slovene): leaderboard
SloBench (CJVT, University of Ljubljana) DIALECT-COPA leaderboard, standard Slovene test set: Choice of Plausible Alternatives commonsense reasoning, picking the more plausible cause or effect of a premise. Accuracy 0-100, higher is better, chance 50; LLM submissions only, prompted through APIs or Ollama.
Metric: Accuracy (%). Source: slobench.cjvt.si. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 99.8 | #105 |
| 2 | Gemini 2.5 Pro | 98.6 | #159 |
| 3 | Gemini 2.5 Flash | 97.4 | #255 |
| 4 | GPT-4o (2024-08-06) | 97 | #343 |
| 5 | Claude Haiku 4.5 | 92.6 | #287 |
| 6 | Mistral Medium 3.1 | 90 | #418 |
| 7 | Llama 3.3 70B Instruct | 87 | #541 |
| 8 | Gemma 3 27B (IT) | 86.2 | #528 |
| 9 | GPT-3.5 Turbo (0125) | 84.2 | #975 |
| 10 | Qwen 3 32B | 82.6 | #449 |
| 11 | DeepSeek R1 Distill Qwen 14B | 53 | #860 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=slobench-dialect-copa-standard-slovene · How It Works · Data refreshed daily, snapshot 2026-11-05.