Llama-3-Lumimaid-8B-v0.1-OAS: benchmark results
A Llama 3 role-play release by NeverSleep, credited to Undi and IkariDev. The card describes a mixture of role-play and general instruction datasets and uses the Llama 3 prompting format. This OAS version received Orthogonal Activation Steering and remains distinct from the original fine-tune. Provider: NeverSleep. Access: Open.
Aggregate Rating 1419 ± 35, rank #5237 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 67.23 | Accuracy (%) (5-shot) | 89 |
| Open LLM Leaderboard v1 - GSM8K | 63.91 | Accuracy (%) (5-shot) | 77.6 |
| Open LLM Leaderboard v1 - ARC Challenge | 61.43 | Normalized accuracy (%) (25-shot) | 52.8 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 49.87 | MC2 (%) (0-shot) | 46.4 |
| Open LLM Leaderboard v1 - WinoGrande | 75.3 | Accuracy (%) (5-shot) | 41.4 |
| Open LLM Leaderboard v1 - HellaSwag | 79.1 | Normalized accuracy (%) (10-shot) | 36.7 |
Interactive version: theaggregate.ai/model?slug=llama-3-lumimaid-8b-v0-1-oas · How It Works · Data refreshed daily, snapshot 2026-11-08.