Registration opens on December 2, 2025. The deadline to register is January 16, 2026. Slots are limited, so register early to ensure your team’s participation. 

The Challenge will kick off on February 16, 2026, with a live welcome call. The Challenge execution takes place in phases through March 2026.

Participants should expect to spend 10-15 hours across 1-2 weeks once the Challenge begins. 

Results will be evaluated and communicated to participants in the second half of March 2026. Winners will be announced shortly after evaluation is complete.

Yes, FeatureByte will host a live informational webinar and interactive FAQ session on December 18, 2025. A recording will be available for interested participants who are unable to attend live. Register here.

Participants will test their production models using real-world relational data—composed of multiple linked tables in their data platforms like Snowflake, Databricks, and BigQuery—not pre-engineered datasets or a single pre-aggregated feature table. FeatureByte will connect into the participants’ data platforms via a service account and perform feature engineering directly on this relational data. For more details on the data and connectivity, please refer to the Challenge Rules.

The Challenge is limited to binary classification models on tabular data. We’ve chosen classification models since they are easier to evaluate. Participants can choose which of their production models they want to test against FeatureByte, as long as the model has been in production for at least 30 days. 

The Challenge is designed to work within controlled, real-world conditions and use cases, rather than pre-engineered data. Participants will need to share the data that was used to build their production model. The data doesn’t need to be live, but should contain all the history needed to build and test the challenger model. Specific data access requirements will be outlined in the onboarding kit sent to registered participants.

We understand that some organizations have strict policies around data privacy due to regulations and compliance issues. If your organization does not allow data to be shared with FeatureByte, unfortunately you will not be able to participate in the Challenge. However, you can still benchmark your models by installing FeatureByte in your virtual private cloud (VPC) and connecting to your data platform. Please contact us at challenge@featurebyte.ai to discuss the process. 

All data will be shared with FeatureByte under a Non-Disclosure Agreement (NDA). The data, use case, model artifacts, etc. will be private to FeatureByte. All data processing (data prep, feature engineering, etc.) will be performed in your data platform. The only datasets that get transmitted to FeatureByte are for training, testing, and validating models. All data will be stored encrypted. Additionally, FeatureByte is SOC 2 Type 2 compliant, with all the relevant safeguards in place for data and access. For more information, download the white paper “Secure by Design: FeatureByte's Approach to Data Privacy & Security.”

We will only request detailed model metrics if your model outperforms FeatureByte’s. Artifacts such as feature importance, feature definitions, description of pre-processing steps, etc., will allow judges to evaluate the model and ensure correctness.   

Winning models will be evaluated by a panel of third-party judges using standardized criteria. The evaluation will check for data leakage, rule compliance, and other performance metrics. 

To simplify model evaluation and benchmarking, we’re using AUC (Area Under the Curve) as the evaluation metric. FeatureByte will also make modeling artifacts such as feature importance, feature definitions, and associated code available to you for further analysis. Learn more about evaluation criteria in the Challenge Rules. 

FeatureByte will provide full, transparent access to results with clear analysis showing where your models excel and where there's improvement potential—no black boxes or blind spots.​ To ensure the judging process is unbiased, FeatureByte will use neutral third-party judges to evaluate models. 

FeatureByte will monitor a dedicated Slack channel and answer participant questions throughout the Challenge. Additionally, we will set up checkpoints with participants to discuss progress and results during the Challenge.

Yes, registered participants will receive onboarding materials in early to mid-February 2026 before the Challenge begins.

Yes, the top three teams whose production models beat FeatureByte’s models (see Challenge Rules) will receive cash prizes.​ First place: $10,000; second place: $5,000; third place: $2,500. If the winning team is unable to accept a monetary award, we will donate the award to a charity of their choice.

Results will be published via post-Challenge blog posts, social media updates, and video content. Winners will be announced and celebrated publicly across multiple channels.​

All participants will receive a recap email with results. FeatureByte may also feature your stories and feedback collected during the Challenge (with your permission, of course).

FeatureByte will conduct follow-ups with teams to discuss findings and potential next steps for improving model performance. The primary contact (Team Leader) of each winning team will receive prize money after the end of the Challenge period. 

The Challenge provides free benchmarking with zero risk. It's an opportunity for business data science teams to validate performance and discover improvement potential without financial commitment.

The Challenge reveals how much value your production models might be leaving on the table, and the additional value that can be delivered through measurable improvements. It can help you uncover hidden inefficiencies in your organization’s data science processes.

For winners, this is a great opportunity to highlight their data science chops both within the organization and outside. Strong data science teams attract more investments and other strong candidates to continue building out the team.