UltraLM-65B [Delta]: benchmark results
A chat fine-tune of the original LLaMA-65B on UltraChat, distributed as delta weights that must be recovered against the base model. The delta release remains distinct from the original LLaMA checkpoint. Provider: OpenBMB. Released 2023-08-07. Access: Open.
Aggregate Rating 1420 ± 34, rank #5165 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - WinoGrande | 81.14 | Accuracy (%) (5-shot) | 77.4 |
| Open LLM Leaderboard v1 - ARC Challenge | 67.06 | Normalized accuracy (%) (25-shot) | 75.6 |
| Open LLM Leaderboard v1 - HellaSwag | 84.98 | Normalized accuracy (%) (10-shot) | 70.6 |
| Open LLM Leaderboard v1 - MMLU | 63.48 | Accuracy (%) (5-shot) | 61 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 53.51 | MC2 (%) (0-shot) | 59.8 |
| Open LLM Leaderboard v1 - GSM8K | 32.75 | Accuracy (%) (5-shot) | 49 |
Interactive version: theaggregate.ai/model?slug=ultralm-65b-delta · How It Works · Data refreshed daily, snapshot 2026-11-08.