archangel_sft-kto_llama13B: benchmark results
Contextual AI Archangel llama13b is a human-feedback aligned release trained with supervised fine-tuning followed by KTO, using SHP, Anthropic HH and OpenAssistant data. It uses the documented TuluV2-style prompt format and remains separate from its base model and other Archangel training objectives. Provider: Contextual AI. Access: Open.
Aggregate Rating 1314 ± 32, rank #8467 of 10916 rated models, from 12 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| RewardBench Prior Sets (0.5 weight) | 57.6 | Score (%) | 47.4 |
| Open LLM Leaderboard v1 - WinoGrande | 76.16 | Accuracy (%) (5-shot) | 46.4 |
| Open LLM Leaderboard v1 - HellaSwag | 80.8 | Normalized accuracy (%) (10-shot) | 42.5 |
| Open LLM Leaderboard v1 - GSM8K | 16.83 | Accuracy (%) (5-shot) | 38.3 |
| Open LLM Leaderboard v1 - ARC Challenge | 56.14 | Normalized accuracy (%) (25-shot) | 34.9 |
| Open LLM Leaderboard v1 - MMLU | 47.84 | Accuracy (%) (5-shot) | 26.8 |
| RewardBench Reasoning | 70.77 | Accuracy (%) | 26 |
| RewardBench Chat | 84.08 | Accuracy (%) | 18.2 |
| RewardBench | 59.52 | Score (%) | 14.2 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 39.42 | MC2 (%) (0-shot) | 11 |
| RewardBench Chat Hard | 37.72 | Accuracy (%) | 8.5 |
| RewardBench Safety | 46.49 | Accuracy (%) | 4 |
Interactive version: theaggregate.ai/model?slug=archangel-sft-kto-llama13b · How It Works · Data refreshed daily, snapshot 2026-11-08.