Anyone can contribute a model output. Here’s how.

Required Fields

Target task: choose the prompt you ran from the prompts list.
Model: specify the model name (e.g., GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro).
Output text: paste the model’s output as-is verbatim, without editing or sanitizing.

(Optional: in the technical details section, you may specify temperature, top-p, and latency if available).

Documentation Levels

The documentation level determines the output’s reliability:

Quick submit (self-report): you submit the output without external evidence — the fastest way to contribute.
Public chat link: you provide a link to the conversation that anyone can open and verify.
API-verified: you submit the output with full API call details. Highest reliability level. To automate this level, mint a personal access token (PAT) at /en/token — it is shown once at creation, and you choose its expiry and can revoke it anytime.

Copy the full output without truncation.
Do not edit or improve the output — the goal is to measure the model as-is.
Specify the model and version precisely; different models have different capabilities.
If you know the temperature, mention it; it affects output diversity.
Use the highest documentation level available to you.