OpenCode helps you work on a project: it can read files, propose changes, write code and run tools. OpenCode is the workspace; the model is the AI you choose; the provider supplies access to it. Installing OpenCode does not include unlimited use of paid models.
What you need and what it costs
Bring a laptop, install OpenCode Desktop (opens in a new tab), and have your course project folder ready. You do not need a powerful graphics card when using a hosted model.
Our proposed starting route is OpenRouter with $10 of prepaid API credit. It lets you change model families inside OpenCode without creating a separate billing account for each one. Reserve another $10 if your project needs it; do not buy it upfront. This $10–20 course allowance is a planning recommendation, not a measured cost or a promise that every project will fit.
OpenRouter charges a credit-purchase fee, currently 5.5% with a $0.80 minimum, so $10 of credits costs $10.80 before any applicable taxes or currency conversion. Leave automatic top-ups disabled. OpenCode itself does not require a paid subscription for this setup.
Start with GPT-6 Luna as the economical candidate. Compare Gemini 3.8 Flash if it struggles. Use expensive models such as Astra selectively, rather than leaving them selected for every edit. These model choices still need our course-specific pilot before being labelled tested recommendations.
The university's Mammouth access is not required. A browser-chat subscription and an API allowance for OpenCode are different products; do not buy a subscription assuming it supplies unlimited OpenCode usage.
Connect your model
- Create an account at OpenRouter (opens in a new tab), then buy credits under Credits (opens in a new tab).
- Create a course API key under Keys (opens in a new tab). Give it a spending limit within your budget. Treat it like a password.
- Open your existing project folder in OpenCode and start a new session. Use the provider connection control to choose OpenRouter and enter the key in its credential form. In the terminal interface, the equivalent is
/connect, then OpenRouter. - Select OpenRouter → GPT-6 Luna in the model selector. In the terminal interface, use
/models. Check the selected model when starting a new session or changing agents. - After your first small task, inspect the changed files and check the actual charge in OpenRouter Activity (opens in a new tab).
Never paste your key into a chat prompt, source file or shared screenshot. A local OpenCode window does not mean the model runs locally: relevant project content is sent to the selected service.
Start with your journey or app
Open the starter folder for your chosen journey, or a folder for the app you want to build. Begin with one useful, checkable step:
Work only inside this project folder. Read the task and supplied files, then explain the smallest useful next step and how I can check it. Do not edit anything yet. Ask if information is missing; do not invent results.
For a guided first build, follow Build Your Own Data Explorer with OpenCode. Use its supplied files and step-by-step checks: experiment A should give 12.00 mm, B 8.00 mm, and All 10.67 mm. Exporting A should contain only S01 and S02. Inspect the actual app and files, not just the agent's claim of success.
Prefer another project? Choose the Document Assistant, Scientific Image Review App, or another journey. Browse Apps for ideas. Keep your own question and begin with one feature: a source-linked document answer or an inspectable image view. An illustrative citation or a displayed detection overlay is not proof that retrieval or image analysis has been implemented. Check the underlying behaviour against your chosen source material.
Ask for a bounded change, inspect it, and give specific feedback. If the model repeatedly fails the same check, stop the loop and compare another model. Keep a short record of the model, outcome and cost.
If you cannot pay
Try a free Gemini API key from Google AI Studio (opens in a new tab). Connect Google in OpenCode and choose a model with a currently available free API tier, such as Gemini 3.8 Flash. Keep billing disabled for a free-only trial. Free-tier quotas can interrupt agent work, and free-tier content may be used to improve Google's products: use public or synthetic course data. Tell the instructor if quota prevents completion.
Evidence and pricing — 1 October 2026: This introduction links to the existing course exercises. The recommended model shortlist has not yet been compared on those exercises, and the five-day budget has not been measured in a classroom. Existing journey recordings do not establish that each model can reproduce them. Prices, availability and free quotas can change; check the provider before purchasing.
Sources: OpenCode V2 (opens in a new tab), provider connections (opens in a new tab), OpenRouter fees (opens in a new tab), live model catalogue (opens in a new tab), Google pricing (opens in a new tab).