Submit Your Reports
Choose your submission tier
Pick the tier that matches your system’s total parameter count*. Each tier has its own evaluation track, so your system is only ever compared against others of comparable size.
* Total parameters, not active ones. A sparse mixture-of-experts model is sized by its full weight count: gpt-oss-120b counts as 117B, not its 5.1B active parameters. Model names are not a reliable guide either — Mistral Small 4 is a 119B-parameter Tier 2 model.
Download the hypothetical patient cases
There are four hypothetical patients. Each case follows the challenge’s input format: an id, a patient.genotype list, and a brief patient.clinical_context.
Generate your reports
- Run your system on each case. Each report must follow the challenge’s output format and the Markdown Report Template — a ready-to-fill template file is available.
- Combine the four output objects into one JSON array, one entry per patient:
[
{
"id": "AITX-40001",
"report_markdown": "# Therapeutic Actionability Report for Patient AITX-40001\n\n## Summary..."
},
{
"id": "AITX-40002",
"report_markdown": "# Therapeutic Actionability Report for Patient AITX-40002\n\n## Summary..."
},
...
]
- Before you can submit, the upload form below checks your file: it must be valid JSON, contain exactly one report for each of the four patient IDs, and include every required report section heading, in order.
Upload your reports
One submission per individual or team. If you need to correct a submission before the deadline, simply submit again; we will evaluate your most recent submission.
Evaluation
- Each report is scored by a panel of experts in clinical genetics, translational research, and drug development, using the challenge’s scoring criteria: factual accuracy, completeness, evidence & justification, prioritization & calibration, and clinical utility.
- Submissions are ranked within their tier, so your system is only compared against others of comparable size.
- Winners will be announced by January 25, 2027.
If you place at the top of your tier
We will contact you to confirm that your reports can be reproduced by your system. Winners will be asked to:
- Build a Hugging Face Space. Implement an
/actionability_reportendpoint that accepts the input format and returns the output format. You may host or call any model, including private and API-based ones, as permitted by your tier. - Set access.
- Public Space: no action needed.
- Private Space: provide a read-only, fine-grained Hugging Face token (Settings → Access Tokens → Create new token) with read access only to the submitted Space repository. Do not share write tokens, or tokens with access to unrelated repositories.
- Register & validate your Space. Sign in at the AI-Tx Challenge submission Space, provide your Space URL and access token (if private), then run the validator to confirm API functionality and response format. Agree to no-logging rules: no logging, storing, redistributing, or training on evaluation queries or responses.
- Pin your base model(s). Confirm the base model(s) named in your submission at an exact version: a repository and revision (commit SHA or tag) for open weights, or the exact model version or snapshot for an API. You do not need to upload a copy of open weights we can already fetch.
- Tier 1 & Tier 2 only Release your artifacts.* Share what is needed to reproduce your system: whatever you built on top of your base model, such as scaffold and inference code, prompts, and any adapter or fine-tuned weights. You can and should still maintain your own licensing.
- If you also did training, provide your training configuration and the training data produced or assembled.
- If you used retrieval, include the index-building code and a manifest of sources with the date of the snapshot.
* When data genuinely cannot be shared (patient consent, third-party licensing), tell us and share what you can: a schema, a small synthetic sample, and the code that consumes it, so the pipeline still runs end to end.
New here? Start with the challenge specifications, the tutorials, and the FAQ.
