bilingual-gpt-neox-4B: benchmark results
Rinna English-Japanese GPT-NeoX model trained on approximately 524B tokens. The publisher reports 3.8B parameters across 36 transformer layers with hidden width 2,816. Provider: rinna. Released 2023-07-31. Access: Open.
Aggregate Rating 1122 ± 16, rank #10485 of 10916 rated models, from 72 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - TruthfulQA MC2 | 45 | Accuracy (%) | 51.3 |
| pfgen-bench - Completion Mode - Helpfulness | 0.09 | Helpfulness Score | 43.8 |
| MMLU-by-task - Machine Learning | 31.25 | Accuracy (%) | 43.7 |
| MMLU-by-task - Formal Logic | 28.57 | Accuracy (%) | 41.7 |
| pfgen-bench - Completion Mode - Truthfulness | 0.66 | Truthfulness Score | 38.7 |
| pfgen-bench - Completion Mode - Score | 0.41 | pfgen Score (mean of three) | 37.7 |
| pfgen-bench - Completion Mode - Fluency | 0.49 | Fluency Score | 31.3 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 45 | MC2 (%) (0-shot) | 29.4 |
| MMLU-by-task - TruthfulQA MC1 | 24.6 | Accuracy (%) | 28.9 |
| MMLU-by-task - World Religions | 32.16 | Accuracy (%) | 28 |
| MMLU-by-task - Business Ethics | 30 | Accuracy (%) | 26.7 |
| MMLU-by-task - Marketing | 29.06 | Accuracy (%) | 26.1 |
Interactive version: theaggregate.ai/model?slug=bilingual-gpt-neox-4b · How It Works · Data refreshed daily, snapshot 2026-11-08.