Starling-LM-7B-beta: benchmark results
Nexusflow's Starling-LM-7B-beta, an RLAIF-trained fine-tune of Openchat-3.5-0106 (Mistral 7B-based), aligned via PPO with the Starling-RM-34B reward model on the Nectar ranking dataset. Provider: Nexusflow. Released 2024-03-20. Access: Open.
Aggregate Rating 1432 ± 12, rank #1093 of 1656 rated models, from 126 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Medical LLM - PubMedQA | 78.2 | Accuracy (%) | 96.1 |
| Open PL LLM - KLEJ NER (generative, 5-shot) | 55.93 | Accuracy (%) | 89.6 |
| Open PL LLM - KLEJ NER (multiple choice, 5-shot) | 55.54 | Accuracy (%) | 86.2 |
| BiGGen-Bench | 3.76 | Average Score (1-5) | 82.4 |
| Open LLM Leaderboard v1 - MMLU | 65.14 | Accuracy (%) (5-shot) | 82.3 |
| Open LLM Leaderboard v1 - GSM8K | 66.64 | Accuracy (%) (5-shot) | 82.1 |
| Open Portuguese LLM - ASSIN2 RTE | 92.57 | Macro F1 (%) | 80.4 |
| Open Portuguese LLM - FaQuAD NLI | 77.49 | Macro F1 (%) | 80.1 |
| Enkrypt AI - Toxicity Risk | 1.52 | Risk Score | 79.6 |
| Open LLM Leaderboard v1 - WinoGrande | 81.29 | Accuracy (%) (5-shot) | 78.3 |
| Open PL LLM - PoQuAD Open Book (generative, 5-shot) | 68.72 | Levenshtein Similarity (%) | 77.7 |
| Open PL LLM - PPC (generative, 5-shot) | 76.8 | Accuracy (%) | 77.4 |
Interactive version: theaggregate.ai/model?slug=starling-lm-7b-beta · How It Works · Data refreshed daily, snapshot 2026-10-20.