Submit

Submit a model.

Two paths: Verified, or Unverified. Verified entries appear above all Unverified entries on the leaderboard, regardless of headline score.

Path 1

Verified (recommended)

Clone the evaluation harness (published under a permissive open license): github.com/blankline-org/joule-index
Run the prep + capture flow against the public Preview tasks using your model's API key.
Publish the sanitized public_trace.json for each run.
Open a PR against the leaderboard repository. The Blankline research team verifies against the source billing record and adds the entry to the table.

Path 2

Unverified

Report the numbers (F1, cost, joules) without the trace.
Get listed in the Unverified section, below all Verified entries.
Anyone watching the leaderboard sees the visibility gap and draws their own conclusion.

Protected

What submissions are not required to expose

Source code of your CLI / agent / scaffolding
System prompts, internal reasoning chains, routing logic
Proprietary model identifiers beyond the public model name
Any internal state that does not bear on what the agent did to the task workspace

Open invitation

To every frontier lab

Blankline has invited Anthropic, OpenAI, Google DeepMind, xAI, Meta, Mistral, DeepSeek, Moonshot, Alibaba, and every other frontier laboratory to submit. The leaderboard publicly tracks which vendors have and which have not.

A model that refuses verification is a model that does not want to be seen.