Task Decomposition on VoxMind Agentic Core Capabilities Dataset
95.24TSVoxMind
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VoxMind2026.04 | 95.24 | 38.1 | |
| Gemini-2.5-proModel Category=Closed-source models, Inference Mode=direct inference with prompt2026.04 | 82.54 | 52.38 | |
| Qwen3-8B+WhisperModel Category=Open-source models, Architecture=Cascaded models, Inference Mode=direct inference with prompt2026.04 | 82.54 | 41.27 | |
| Gemini-2.5-flashModel Category=Closed-source models, Inference Mode=direct inference with prompt2026.04 | 61.9 | 31.22 | |
| StepAudio2Model Category=Open-source models, Architecture=End-to-end models, Inference Mode=direct inference with prompt2026.04 | 60.32 | 26.98 | |
| Kimi-AudioModel Category=Open-source models, Architecture=End-to-end models, Inference Mode=direct inference with prompt2026.04 | 48.15 | 22.75 | |
| Qwen2.5-OmniModel Category=Open-source models, Architecture=End-to-end models, Inference Mode=direct inference with prompt2026.04 | 38.62 | 3.17 | |
| GPT-4o-audioModel Category=Closed-source models, Inference Mode=direct inference with prompt2026.04 | 23.81 | 15.87 |