Skip to content

Benchmark

IdrockBench Uzbek core composite (chance-normalised; dtm 2.1, reasoning_uz 2.0, translation_uz 2.0, ifeval_uz 2.0, zarbulmasal 1.0)

Instruction following benchmarks measure how closely models follow instructions. Scores are index points out of 100. Higher is better.

Models measured
12
Scores recorded
12
Sources
1
IQ weight
0.75

Score distribution

31.5
Mean
30.7
Median
15.4
Std dev
12.9
10th percentile
52
90th percentile
9.2
Min
9.252.3
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.75 for chat models, measured over 12 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Qwen3.6 27BQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs52.3
2Gemma 4 31B ITQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs52
3Qwen3.5 27BQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs48.5
4Qwen 3.8 27BQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs45.7
5Gemma 4 26BQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs42.2
6Qwen3.5 9BQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs30.7
7gemma 4 E4BQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs29.5
8Qwen3.5 4BQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs22.4
9Gemma 4 E2B ITQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs20
10Qwen3.5 2BQ8_0, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs13.2
11Nemotron 3.5 Lightning 30bQ4_K_M, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs12.9
12Qwen3.5 0.8BQ8_0, reasoning: no-think, organiser-run· IdrockBench Uzbek leaderboard data docs9.2

Data credits: IdrockBench, IDROCK AI Excellence Center, New Uzbekistan University. Public numeric results; no results licence stated (code MIT and dataset licences not borrowed).