Open-ended Question Answering on Proposed LLM-based evaluation benchmark OEQ
96.9CompletenessGPT-4o-Mini-Audio
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4o-Mini-AudioModel Configuration=GPT-4o-Mini-Audio2025.12 | 96.9 | 49 | 98.1 | 68.4 | |
| GPT-4o-AudioModel Configuration=GPT-4o-Audio2025.12 | 96.8 | 47.4 | 97.2 | 67.2 | |
| Gemini-2.5-FlashModel Configuration=Gemini-2.5-Flash2025.12 | 89.1 | 45.2 | 88.9 | 62.7 | |
| Qwen2.5-Omni-7BModel Configuration=Qwen2.5-Omni-7B2025.12 | 77.8 | 62.2 | 76.9 | 68.3 | |
| Gemini-2.5-ProModel Configuration=Gemini-2.5-Pro2025.12 | 77.7 | 28.4 | 76.9 | 48 | |
| SFT with MLLM-only dataModel Configuration=SFT with MLLM-only data, Backbone=Qwen2.5-Omni-7B2025.12 | 77.7 | 73.5 | 71.3 | 73.9 | |
| SFT with Hybrid dataModel Configuration=SFT with Hybrid data, Backbone=Qwen2.5-Omni-7B2025.12 | 77 | 51.5 | 56.9 | 57.7 | |
| Qwen2-Audio-7BModel Configuration=Qwen2-Audio-7B2025.12 | 68.5 | 44.8 | 48.1 | 50.2 | |
| SFT with Human-only dataModel Configuration=SFT with Human-only data, Backbone=Qwen2.5-Omni-7B2025.12 | 55.9 | 24.7 | 57.4 | 37.5 |