PLLuM-8x7B-chat: benchmark results
CYFRA/PLLuM's PLLuM-8x7B-chat, a Mixtral-style mixture-of-experts chat model specialized for Polish and other Slavic/Baltic languages, instruction-tuned on a curated Polish 'organic instructions' dataset. Provider: PLLuM. Released 2025-02-24. Access: Open.
Aggregate Rating 1418 ± 20, rank #1136 of 1635 rated models, from 19 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLMZSZL Leaderboard | 52.8 | Score | 68.4 |
| MT-Bench PL - Extraction | 8 | Judge Score (0-10) | 42.9 |
| MT-Bench PL - Overall | 6.3 | Judge Score (0-10) | 40.8 |
| MT-Bench PL - STEM | 8.2 | Judge Score (0-10) | 39.8 |
| MT-Bench PL - Reasoning | 4.9 | Judge Score (0-10) | 36.7 |
| MT-Bench PL - Coding | 4.55 | Judge Score (0-10) | 35.7 |
| PLCC - Culture & Tradition | 60 | Accuracy (%) | 35.6 |
| MT-Bench PL - Humanities | 8.6 | Judge Score (0-10) | 34.7 |
| CPTU Bench | 3.01 | Average Score (1-5) | 31.7 |
| PLCC - Art & Entertainment | 45 | Accuracy (%) | 31.7 |
| Polish EQ-Bench | 45.22 | EQ-Bench Score (parseability-adjusted) | 31.7 |
| PLCC - History | 68 | Accuracy (%) | 31.3 |
Interactive version: theaggregate.ai/model?slug=pllum-8x7b-chat · How It Works · Data refreshed daily, snapshot 2026-11-05.