Q1: How to determine whether an open-source model or an open-source data can be used?
A: The release cut-off for the model and data is November 30, 2025. If you want to use LLM or MLLM to label the data, you need to make sure that your audio comes from an open-source dataset and that the annotation model is also open-source released before November 30, 2025.

Q1: How to submit results to the leaderboard?
A: The leaderboard will be available starting from January 1st, 2026. You can train your single model or design your agent before this point. Only teams that register before January 15, 2026 can submit their results.

Q3: Will the paper be accepted by the main conference?
A: Yes, and submissions describing the competition systems or reporting research results based on the competition benchmark are equally welcome. Besides, the top-performing teams will be awarded with honors in the conference.

Q4: How is the final rank determined? A: Your highest score from the Final Stage will be used for the final ranking. The leaderboard updates in real-time with about 20 minutes delay.

Q5: Can I use local database or knowledge base for the Agent Track? A: No, external knowledge bases are not allowed. This may lead to unfair advantages and potential data leakage.

Q6: Can I use other meta information, such as the category or sub-category in the meta json file for reasoning? A: No, only audio, question, and choices, along with a input prompt are allowed as input.