AISE-Bench Leaderboard

Evaluating academic information-seeking agents across planning, API execution, reference grounding, and answer quality.

500official test questions
10 metricsprocess and answer evaluation
Verified onlyscores produced by the official evaluator

How to Submit

  1. Fork this repository.
  2. Create a new branch for your submission.
  3. Add your submission folder under submissions/<organization>__<model>__<input_config>/.
  4. Open a Pull Request with the new submission folder.

Submission Directory Requirements

Each submission directory must contain the metadata and predictions for one model/input configuration pair:

<organization>__<model>__<input_config>/
  metadata.yaml
  predictions.jsonl
  generation_config.json      # optional, recommended
  artifacts/                  # optional logs or prompt notes

Use URL-safe directory names. Replace spaces, slashes, and special characters with hyphens; keep input_config as TEXT.

metadata.yaml

model_name: "My Model"
model_url: https://...

predictions.jsonl

Each line must be one JSON object:

{
  "index": 1,
  "qid": 3,
  "question": "question text",
  "planning_text": [],
  "api_input": {},
  "api_output": {},
  "result_edit": "{\"answer\":\"...\",\"reference\":{\"[1]\":\"...\"}}"
}

The file must contain exactly one line for every item in test.json. qid must match the corresponding item in test.json.

Leaderboard

Official results on the 500-question test set are sorted by F1-LM by default. The table is intentionally empty until the first submission completes official evaluation and review.

Rank Model Track Planning Exec. Precision Recall Format Edit Dist. Para. Acc. Success Correct. Complete. Faithful. F1-LM
No accepted submissions yet.