Is qwen3-plus Required as the JSON Conversion Model in DeepPlanning Benchmarks?

[fjcobu14]

In DeepPlanning, does the model that converts the model’s response into a JSON structure have to be qwen3-plus?
For the leaderboard scores you reported, was the conversion model also set to qwen3-plus?

[zhangyingerjelly]

Hi, thank you for your interest in DeepPlanning.

Regarding your questions:

Leaderboard Consistency: Yes, all scores currently reported on our leaderboard were generated using qwen-plus (or your specific version) as the conversion model. To ensure a fair and direct comparison with these benchmarks, we recommend using the same configuration.
Model Flexibility: While we used qwen-plus for the baseline, it is not strictly required. Our experiments indicate that the error rate in this specific conversion step (mapping responses to JSON) is quite low. Therefore, if you have resource constraints or prefer another model, you can substitute it with an alternative model without significantly impacting the results.