OpenRouter

OpenRouter is a unified API gateway that gives you access to hundreds of AI models — from providers like Anthropic, Google, Meta, Mistral, DeepSeek, and more — through a single API key and a single billing account. Instead of signing up separately with each AI provider and managing multiple API keys, you create one OpenRouter account, fund it with a prepaid balance, and then specify which model you want to use at request time.

OpenRouter is the only supported provider in GrillMyCode, and the default — you can omit ai_provider entirely. Because OpenRouter is a gateway rather than a single vendor, you still choose freely between models from Anthropic, Google, DeepSeek, Meta, Mistral, and others — the choice is made in ai_model, not by swapping providers.

An API key is required

OpenRouter cannot authenticate with the built-in GITHUB_TOKEN. Every GrillMyCode workflow must supply api_key. The action fails immediately with a setup message if it is missing.

Migrating a workflow that used to work and no longer does? See the FAQ.

Instructor setup guide

To use OpenRouter with GrillMyCode, an instructor needs to perform the following steps once:

1. Create an OpenRouter account — Go to openrouter.ai and sign up (Google, GitHub, or email).
2. Add credit — Navigate to openrouter.ai/credits and add a prepaid balance. Many of the recommended models below cost fractions of a cent per call, so $5–10 will last a large class for an entire semester.
3. Generate an API key — Go to openrouter.ai/keys and create a new key. Copy it immediately — you won't be able to view it again.
4. Store the key as an organisation-level GitHub secret — Go to your GitHub organisation's Settings → Secrets and variables → Actions and create a new secret named OPENROUTER_API_KEY with the key from Step 3. This makes the key available to all student repositories for that classroom automatically without any per-repo configuration.
5. Configure the workflow — Pass the key as api_key and choose a model via ai_model (see the tables below for recommended options). ai_provider can be omitted, since openrouter is the default.

Classroom tip

For a Classroom 50 setup the secret should always be added at the organization level. This ensures every student repo has access to the key from the moment it is created, with no extra setup required from students.

Required secrets

Secret name	Description
OPENROUTER_API_KEY	Your OpenRouter API key

Inputs

Input	Value
ai_provider	openrouter (the default — may be omitted)
api_key	${{ secrets.OPENROUTER_API_KEY }} — required
ai_model	Any model identifier supported by OpenRouter, in provider/model-name format. Defaults to google/gemini-3.5-flash-lite. See the OpenRouter model list. Examples: openai/gpt-4o, anthropic/claude-3-5-sonnet, meta-llama/llama-3.1-70b-instruct

Recommended models

The following models have been tested with GrillMyCode and are all usually under 1 cent per API call at typical classroom scale. They are the options pre-loaded in the workflow wizard.

Model	ai_model value	Notes
Google Gemini 3.5 Flash Lite ⭐ Recommended	google/gemini-3.5-flash-lite (default)	Google's entry-level flash tier; fast, cheap, and consistent for short-form generation tasks. Of the tested models it produces the most effective distractors for multiple-choice questions, which is why it is both recommended and the default.
DeepSeek V4 Flash	deepseek/deepseek-v4-flash	Extremely low cost; strong instruction-following for structured JSON output. A reliable alternative.
Minimax M2.7	minimax/minimax-m2.7	Very inexpensive; performs well on question generation with minimal prompt tuning.
StepFun Step 3.5 Flash	stepfun/step-3.5-flash	Competitive quality-per-token ratio; tested to produce well-formed assessment questions.
Tencent Hy3	tencent/hy3	Model from Tencent; cheap and functional, though output style may vary.
Xiaomi MiMo V2.5 Pro	xiaomi/mimo-v2.5-pro	Reasoning-optimised model from Xiaomi; good at following structured output constraints.

All six are in the "free-or-near-free" tier on OpenRouter, making them suitable for deployments where many students submit simultaneously. If you want higher output quality and are willing to pay more, you can use any other OpenRouter model via the Own Choice option — just verify pricing at openrouter.ai/models first.

Example

- uses: NSCC-ITC-Assessment/GrillMyCode@v1
  with:
    github_token: ${{ secrets.GITHUB_TOKEN }}
    api_key: ${{ secrets.OPENROUTER_API_KEY }}
    ai_model: 'google/gemini-3.5-flash-lite'

To pin the provider explicitly — harmless, but unnecessary — add ai_provider: 'openrouter'.