Define custom instructions that are automatically appended to every question generation request. Use Markdown and LaTeX freely.
Choose which set of system prompts to use for question generation. Leave on "Server default" unless you want to A/B test a specific version.
Choose which AI model to use for question generation. "System default" uses GPT-4o normally and o3-mini when Think Hard is enabled. When you pick a specific model, Think Hard keeps that same model and raises its reasoning effort instead — this applies to OpenAI reasoning models only (o3, o3-mini, o4-mini, GPT-5.6 family), and has no effect on GPT-4o or Claude models.
ANTHROPIC_API_KEY to be set on the server.
OpenAI models require OPENAI_API_KEY.
This instruction is added to every Math question generation request.
This instruction is added to every Code question generation request.