dolphin-2.9.2-Phi-3-Medium-abliterated: benchmark results
Cognitive Computations' Dolphin 2.9.2 tune of Phi-3 Medium 14B with refusals further ablated by FailSpy, for uncensored instruction, chat, and coding use. Provider: Cognitive Computations. Released 2024-06-03. Access: Open.
Unified ELO 1471 ± 1, rank #835 of 1392 rated models, from 15 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - BBH | 48.39 | Score | 88.5 |
| Open LLM Leaderboard - MMLU-Pro | 39.16 | Score | 84.2 |
| Open LLM Leaderboard - GPQA | 10.51 | Score | 80.1 |
| Open Chinese LLM - GSM8K | 56.33 | Accuracy (%) | 77.9 |
| Open LLM Leaderboard - MuSR | 13.73 | Score | 75.5 |
| Open LLM Leaderboard - MATH Level 5 | 18.2 | Score | 67.8 |
| Open Chinese LLM - C-Eval Semantic | 78.18 | Accuracy (%) | 67.7 |
| Open Chinese LLM Leaderboard | 59.12 | Average Score (%) | 67.4 |
| Open Chinese LLM - CMMLU | 58.03 | Accuracy (%) | 65.6 |
| Open Chinese LLM - WinoGrande | 63.61 | Accuracy (%) | 59.2 |
| Open Korean LLM Leaderboard | 36.04 | Average Score (%) | 52.2 |
| Open Chinese LLM - ARC Challenge | 51.28 | Accuracy (%) | 45.7 |
Interactive version: theaggregate.ai/model?slug=dolphin-2-9-2-phi-3-medium-abliterated · How It Works · Data refreshed daily, snapshot 2026-09-05.