Update Oct 10, 2024

<aside> 💡

With the release of our new open-source project, we had to make minor changes to the API, so now the URL you make requests to, needs to have a trail /

</aside>

This page documents the process for getting and using OpenAI and Anthropic credits for use in the Q4 AI Forecasting Benchmark tournament.

We hope to support both new and experienced bot builders. We will grant between $100 and $500 in credits initially depending on your description and needs. We have the capacity to grant more to builders who demonstrate good use of the initial credits and where we believe that more credits could be used productively.

Step 1: Make a request

If you’re interested in credits, please send an email to [email protected] with:

• A couple paragraph description of your existing bot, or what you would like to build.
• The name of your bot.
• An estimate of how much in $ you expect you could productively use.
• Our default will be to issue credits for GPT4o, but we can distribute the credits between models at your request.

We will aim to review your request within 1 business day.

Technical details

We proxy calls to OpenAI and Anthropic so we can count the tokens usage per user in order to allow a quota of credits for each user. So instead of making a request to OpenAI or Anthropic, you will make a request to our platform, authenticated with your Metaculus Token that you’re already using to predict on questions, and then we forward it to the respective platform and return to you their response. Before you can do that, you have to write us and request an allowance for credits for a particular platform and model. Because different models have different costs, we need to keep track of each model usage.

Anthropic

We can proxy for you the requests to Anthropic’s messages endpoint.

So instead of making a request to Anthropic, like they say in their doc: