Hallucination Detection on BRACE Hallucination 1.0 (test)
98.8AudioCaps ScoreAQAScore
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AQAScoreModel=AF3-Think2026.01 | 98.8 | 98 | — | |
| AQAScoreModel=Qwen2.5-Omni-7B2026.01 | 98.7 | 98 | — | |
| FLEURBackbone=Qwen3-Omni-Instruct, Metric Architecture=LALM-ONLY2026.03 | 98.48 | 98 | 98.18 | |
| AQAScoreModel=Qwen2.5-Omni-3B2026.01 | 98.3 | 97.1 | — | |
| CAF-ScoreBackbone LALM=Qwen3-Omni-Instruct, Backbone CLAP=M2D-CLAP, Metric Architecture=CLAP + LALM2026.03 | 97.96 | 96.59 | 97.11 | |
| CAF-ScoreBackbone LALM=Qwen3-Omni-Instruct, Backbone CLAP=MGA-CLAP, Metric Architecture=CLAP + LALM2026.03 | 97.57 | 96.8 | 97.09 | |
| PromptingModel=Qwen2.5-Omni-7B2026.01 | 97.2 | 97.7 | — | |
| AQAScoreModel=AF32026.01 | 96.9 | 96.9 | — | |
| FENSEReference Protocol=Reference-based2026.03 | 96.76 | 96.18 | 96.4 | |
| CAF-ScoreBackbone LALM=Qwen3-Omni-Instruct, Backbone CLAP=LAION-CLAP, Metric Architecture=CLAP + LALM2026.03 | 96.24 | 94.94 | 95.43 | |
| CAF-ScoreBackbone LALM=AudioFlamingo3-Base, Backbone CLAP=MGA-CLAP, Metric Architecture=CLAP + LALM2026.03 | 95.84 | 93.2 | 94.2 | |
| CAF-ScoreBackbone LALM=AudioFlamingo3-Base, Backbone CLAP=M2D-CLAP, Metric Architecture=CLAP + LALM2026.03 | 95.69 | 92.38 | 93.63 | |
| CascadeModel=Qwen2.5-Omni-7B2026.01 | 95.4 | 95.9 | — | |
| PromptingModel=Qwen2.5-Omni-3B2026.01 | 94.9 | 95.4 | — | |
| AQAScoreModel=AF3-Chat2026.01 | 94.6 | 94.5 | — | |
| FLEURBackbone=AudioFlamingo3-Base, Metric Architecture=LALM-ONLY2026.03 | 93.83 | 95.86 | 95.09 | |
| CAF-ScoreBackbone LALM=Qwen3-Omni-Instruct, Backbone CLAP=MS-CLAP, Metric Architecture=CLAP + LALM2026.03 | 93.52 | 96.61 | 95.44 | |
| CAF-ScoreBackbone LALM=AudioFlamingo3-Base, Backbone CLAP=LAION-CLAP, Metric Architecture=CLAP + LALM2026.03 | 93.2 | 89.98 | 91.2 | |
| CLAPScoreBackbone=MGA-CLAP, Metric Architecture=CLAP-ONLY2026.03 | 92.18 | 87.26 | 89.12 | |
| S-CLAPScoreBackbone=MGA-CLAP, Metric Architecture=CLAP-ONLY2026.03 | 92.18 | 88.08 | 89.48 | |
| CLAPScoreModel=630k-best2026.01 | 91.8 | 82.2 | — | |
| CLAIR-AReference Protocol=Reference-based2026.03 | 91.45 | 91.55 | 91.51 | |
| CascadeModel=Qwen2.5-Omni-3B2026.01 | 91.2 | 90.2 | — | |
| CLAPScoreBackbone=M2D-CLAP, Metric Architecture=CLAP-ONLY2026.03 | 91.03 | 83.93 | 86.61 | |
| S-CLAPScoreBackbone=M2D-CLAP, Metric Architecture=CLAP-ONLY2026.03 | 91.03 | 84.47 | 86.98 | |
| CLAPScoreModel=Music+AS2026.01 | 90.8 | 82 | — | |
| CascadeModel=AF3-Chat2026.01 | 90.5 | 85.1 | — | |
| CLAPScoreModel=Music+Speech+AS2026.01 | 90 | 81.7 | — | |
| FLEURBackbone=Qwen3-Omni-Thinking, Metric Architecture=LALM-ONLY2026.03 | 89.7 | 86.99 | 88.01 | |
| CLAPScoreModel=Music+Speech2026.01 | 88.7 | 81.7 | — | |
| CAF-ScoreBackbone LALM=AudioFlamingo3-Base, Backbone CLAP=MS-CLAP, Metric Architecture=CLAP + LALM2026.03 | 88.16 | 92.98 | 91.16 | |
| CLAPScoreModel=630k-AS2026.01 | 87.6 | 78 | — | |
| CascadeModel=AF32026.01 | 87.5 | 85 | — | |
| PromptingModel=AF3-Chat2026.01 | 87.1 | 85.6 | — | |
| CLAPScoreBackbone=LAION-CLAP, Metric Architecture=CLAP-ONLY2026.03 | 87.03 | 84.31 | 85.34 | |
| S-CLAPScoreBackbone=LAION-CLAP, Metric Architecture=CLAP-ONLY2026.03 | 87.03 | 84.45 | 85.12 | |
| FLEURBackbone=AudioFlamingo3-Think, Metric Architecture=LALM-ONLY2026.03 | 85.8 | 90.14 | 88.5 | |
| PromptingModel=AF32026.01 | 82.3 | 83 | — | |
| Raw ScoreBackbone=Qwen3-Omni-Thinking, Metric Architecture=LALM-ONLY2026.03 | 81.86 | 75.8 | 78.09 | |
| Raw ScoreBackbone=Qwen3-Omni-Instruct, Metric Architecture=LALM-ONLY2026.03 | 81.78 | 82.47 | 82.21 | |
| S-CLAPScoreBackbone=MS-CLAP, Metric Architecture=CLAP-ONLY2026.03 | 73.73 | 87.07 | 82.04 | |
| CLAPScoreBackbone=MS-CLAP, Metric Architecture=CLAP-ONLY2026.03 | 72.99 | 85.23 | 80.61 | |
| CascadeModel=AF3-Think2026.01 | 67.5 | 74.2 | — | |
| PromptingModel=AF3-Think2026.01 | 61.2 | 85.1 | — | |
| Raw ScoreBackbone=AudioFlamingo3-Base, Metric Architecture=LALM-ONLY2026.03 | 25.28 | 60.83 | 47.41 | |
| Raw ScoreBackbone=AudioFlamingo3-Think, Metric Architecture=LALM-ONLY2026.03 | 24.81 | 21.9 | 23 |