SloBench - DIALECT-COPA (Cerkno Dialect): leaderboard

SloBench (CJVT, University of Ljubljana) DIALECT-COPA leaderboard, dialect test set: the COPA commonsense reasoning items written in the Idrija dialect of the Cerkno dialect group, measuring reasoning in non-standard Slovene. Accuracy 0-100, higher is better, chance 50; LLM submissions only.

Metric: Accuracy (%). Source: slobench.cjvt.si. Saturation forecast: Around February 2027. 14 models tracked.

Top models

#ModelScoreOverall rank
1GPT-588.6#105
2Gemini 2.5 Pro86.4#159
3Gemini 2.5 Flash74.2#255
4GPT-4o (2024-08-06)67.6#343
5Gemma 3 27B (IT)57.8#528
6Claude Haiku 4.556.2#287
7Qwen 3 32B54.4#449
8Llama 3.3 70B Instruct53.6#541
9Mistral Medium 3.153.2#418
10GPT-3.5 Turbo (0125)52.8#975
11DeepSeek R1 Distill Qwen 14B49.2#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-cerkno-dialect · How It Works · Data refreshed daily, snapshot 2026-11-05.