EuroEval Serbian NLU - ScaLA SR: leaderboard

The "ScaLA SR" subtask of EuroEval Serbian NLU. Serbian-language EuroEval natural-language-understanding suite, separating NLU task performance from the broader all-task EuroEval aggregate.

Metric: Matthews correlation coefficient (%). Source: euroeval.com. Saturation forecast: Estimated already saturated. 205 models tracked.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro (Preview)81.14#66
2Gemini 3.1 Flash Lite (Preview)79.48#194
3Gemini 3 Flash (Preview) (Non-reasoning)78.24#82 (Gemini 3 Flash (Preview))
4GPT-5.275.24#110
5Gemini 2.5 Flash (Thinking)74.36#245 (Gemini 2.5 Flash)
6Gemini 3 Flash (Preview) (Thinking)72.98#82 (Gemini 3 Flash (Preview))
7GPT-5.4 Mini (2026-03-17) (High)72.8#169 (GPT-5.4 Mini (2026-03-17))
8GPT-572.57#98
9Gemini 2.5 Pro71.6#152
10Claude Sonnet 4.668.97#90
11Gemma 4 31B (IT)68.75#176
12GPT-4.167.6#248
13GPT-5 Mini (2025-08-07)67.09#172
14Gemini 2.5 Flash (Non-reasoning)66.5#245 (Gemini 2.5 Flash)
15Claude Sonnet 4.5 (Thinking)66.43#142 (Claude Sonnet 4.5)

No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.

Interactive version: theaggregate.ai/benchmark?slug=euroeval-serbian-nlu-scala-sr · How It Works · Data refreshed daily, snapshot 2026-10-20.