Skip to content

Benchmark

KazBench Kazakh instruction_following (judge, dev)

Instruction following benchmarks measure how closely models follow instructions. Higher scores are better.

Models measured
6
Scores recorded
6
Sources
1
IQ weight
0.00

Score distribution

0.519
Mean
0.576
Median
0.189
Std dev
0.35
10th percentile
0.807
90th percentile
0.221
Min
0.2210.807
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.00 for chat models, measured over 6 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Haiku 3.5dev, version: 0.1.0· KazBench Kazakh data docs0.807
2Llama 4 Scoutdev, version: 0.1.0· KazBench Kazakh data docs0.62
3Llama 3.1 8B Instantdev, version: 0.1.0· KazBench Kazakh data docs0.576
4GPT-4o minidev, version: 0.1.0· KazBench Kazakh data docs0.538
5Gemini 2.5 Flashdev, version: 0.1.0· KazBench Kazakh data docs0.35
6Qwen2.5 72B Instructdev, version: 0.1.0· KazBench Kazakh data docs0.221

Data credits: KazBench by Yersultan Akhmer. Provisional DEV numeric results; code MIT and task-data CC BY 4.0 do not establish a separate score licence.