nucleus-22B-token-500B: benchmark results
Nucleus.AI's 22B-parameter causal decoder-only base model, pretrained on 500B tokens of RefinedWeb plus curated corpora, meant for further fine-tuning. Provider: NucleusAI. Released 2023-10-06. Access: Open.
Aggregate Rating 1172 ± 31, rank #1585 of 1635 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - WinoGrande | 67.64 | Accuracy (%) (5-shot) | 21.6 |
| Open Portuguese LLM - OAB Exams | 26.92 | Accuracy (%) | 20.7 |
| Open LLM Leaderboard v1 - HellaSwag | 69.39 | Normalized accuracy (%) (10-shot) | 20 |
| Open Portuguese LLM - FaQuAD NLI | 43.97 | Macro F1 (%) | 17.1 |
| Open LLM Leaderboard v1 - ARC Challenge | 40.7 | Normalized accuracy (%) (25-shot) | 17 |
| Open LLM Leaderboard v1 - MMLU | 30.11 | Accuracy (%) (5-shot) | 16.6 |
| Open Portuguese LLM - BLUEX | 23.23 | Accuracy (%) | 16.2 |
| Open Portuguese LLM - ENEM | 20.78 | Accuracy (%) | 15.1 |
| Open LLM Leaderboard v1 - GSM8K | 0.99 | Accuracy (%) (5-shot) | 15 |
| Open LLM Leaderboard - MuSR | 35.11 | Score | 13 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 39.16 | MC2 (%) (0-shot) | 10.3 |
| Open LLM Leaderboard - MMLU-Pro | 11.62 | Score | 7.5 |
Interactive version: theaggregate.ai/model?slug=nucleus-22b-token-500b · How It Works · Data refreshed daily, snapshot 2026-11-05.