ultra0: benchmark results
liminerity's ultra0 is a DARE-TIES merge of starsnatched/MemGPT-2B, liminerity/binarized-ingotrix-slerp-7b. The released configuration uses 24 mistral decoder layers; merge recipe dtype is not a measured evaluator precision. Provider: liminerity. Access: Open.
Aggregate Rating 1223 ± 36, rank #9462 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 16.07 | Accuracy (%) (5-shot) | 37.9 |
| Open LLM Leaderboard v1 - HellaSwag | 68.02 | Normalized accuracy (%) (10-shot) | 19 |
| Open LLM Leaderboard v1 - WinoGrande | 65.51 | Accuracy (%) (5-shot) | 19 |
| Open LLM Leaderboard v1 - MMLU | 33.37 | Accuracy (%) (5-shot) | 17.7 |
| Open LLM Leaderboard v1 - ARC Challenge | 41.47 | Normalized accuracy (%) (25-shot) | 17.6 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 41.49 | MC2 (%) (0-shot) | 17.2 |
Interactive version: theaggregate.ai/model?slug=ultra0 · How It Works · Data refreshed daily, snapshot 2026-11-08.