# Model access for a class
URL: https://research-memex.org/docs/case-studies/systematic-reviews/model-access-for-a-class
Description: Three ways a course gets its participants to a model: one gateway with a capped key each, cloud credits for class projects, or the access students already hold.
For instructors · 15 min reading · no account needed
*Decide how the class reaches a model before the first session, not during it.*
A class needs model access for Sessions 2 to 4 of the [course](/docs/case-studies/systematic-reviews/systematic-review-syllabus). Three routes cover most courses. Pick by what the class will do: compare frontier models on the same task, build and run an application, or both. Reading, peer work and inspecting a supplied response need no model access at all; Session 1 is deliberately done by hand.
| The class will | Start with |
| ---------------------------------------------------- | ---------------------------------------------------------------------------- |
| Run the same task through several models and compare | [One gateway, one key per participant](#one-gateway-one-key-per-participant) |
| Build or host an application that calls a model | [Cloud credits for class projects](#cloud-credits-for-class-projects) |
| Mostly read, discuss and check supplied outputs | [What participants already hold](#what-participants-already-hold) |
Whichever route you choose, test it end to end a week before the cohort starts: one participant account, one request, one look at the dashboard that shows who used what.
## What participants already hold [#what-participants-already-hold]
Exhaust what is free before buying anything. Google AI Studio's free tier serves the course exercises, and a participant with a paid chat subscription may already hold an agent-capable model. The [API Keys guide](/docs/implementation/ai-environment-setup/api-keys-setup-guide) lists the free tiers; the [AI Model Reference Guide](/docs/implementation/core-references/ai-model-reference-guide#free-and-low-cost-options) covers free and local options.
The limit is uniformity. Free tiers rate-limit, and a class on mixed subscriptions cannot run one task through the same frontier models and compare. When the comparison is the point of the session, move to a gateway.
## One gateway, one key per participant [#one-gateway-one-key-per-participant]
This is the arrangement the course has run twice. One paid account on a gateway; buy credits once; issue each participant a named key with its own spend cap. Nobody handles payment, a capped key cannot overspend the budget, and one dashboard shows who used what. The paid credits exist so that everyone can run the same task through different frontier models and feel the difference.
The [API Keys guide](/docs/implementation/ai-environment-setup/api-keys-setup-guide#provider-setup) has the per-key steps for OpenRouter, the gateway the first run used, and names the alternatives: Haimaker, which the second run moved to after OpenRouter credit problems, plus Vercel and Cloudflare gateways. Participants then paste their key into [Cherry Studio](/docs/implementation/agentic-ai-tools/cherry-studio-setup-guide#configure-api-provider) and need no provider account of their own.
Budget for the course's tasks, not for open-ended use. Set each key's cap, watch the dashboard after the first session, and disable keys when the course ends. A participant who wants to continue can open their own account; the course key is not theirs to keep.
## Cloud credits for class projects [#cloud-credits-for-class-projects]
Cloud education programs suit a class that builds something: a prototype that calls a model, an application with storage and hosting, or a data pipeline. They are a poor fit for a class that only needs a chat model; use a gateway for that.
Some programs help students write software. Others pay for the models and cloud services that software uses. For a class building AI applications, you may want both: a coding agent to develop the application, and cloud credits to run it. Each program's eligible services matter more than the provider's model catalog; check coverage for the models your project needs before you design the assignment around them.
### Build applications with a coding agent [#build-applications-with-a-coding-agent]
GitHub Copilot is a useful starting point for a course where students build a website, prototype an application or learn to work with coding agents. Copilot can help implement features, fix bugs and write tests. Its [agent tools](https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent) go beyond suggesting the next line of code.
Verified teachers and students can get free Copilot benefits through GitHub Education. Follow the separate [teacher](https://docs.github.com/en/copilot/how-tos/copilot-on-github/set-up-copilot/enable-copilot/set-up-for-teachers-and-os-maintainers) and [student](https://docs.github.com/en/copilot/how-tos/copilot-on-github/set-up-copilot/enable-copilot/set-up-for-students) routes for the included features and usage allowances.
This supports the work of building an application. It does not pay for external model APIs or hosting that the finished application uses.
### Run a class project on Google Cloud [#run-a-class-project-on-google-cloud]
Google Cloud teaching credits suit a course where students experiment with model APIs, build AI workflows or host applications with storage and other cloud services. Google's teaching-credit route starts with a faculty application.
The [application](https://edu.google.com/programs/credits/teaching/?modal_active=none) asks what the course will do with Google Cloud and how many students will participate. See the [eligibility rules](https://support.google.com/google-cloud-higher-ed/answer/10723190?hl=en) for supported institutions and countries.
Have an award already? [Redeem it](#redeem-a-google-award). With an assigned Cloud project, go straight to [Vertex AI setup](/docs/implementation/ai-environment-setup/vertex-ai-setup#check-your-project), then connect a desktop client or your own application.
### Give students room to experiment on Azure [#give-students-room-to-experiment-on-azure]
[Azure for Students](https://azure.microsoft.com/en-us/free/students/) is an individual route: eligible students sign up for their own credits, with no credit card required. It is worth exploring for independent projects and capstones that combine AI, data and a hosted application, especially when students want to choose what to build rather than follow a shared lab.
The offer excludes third-party and Marketplace products. The [offer terms](https://azure.microsoft.com/en-us/pricing/offers/ms-azr-0170p/) explain eligibility and coverage; a model being available somewhere in Azure does not mean this student offer funds it.
### Teach cloud deployment in managed labs [#teach-cloud-deployment-in-managed-labs]
[AWS Academy Learner Lab](https://aws.amazon.com/training/awsacademy/) provides managed environments for your own assignments, with selected AWS services and usage monitoring. Educators use it through an [AWS Academy member institution](https://aws.amazon.com/training/awsacademy/faq/).
This is most relevant when cloud infrastructure is part of the learning: deploying an application, connecting services or building a data pipeline. For a class that only needs access to a chat model, it is a less direct fit. Check the lab's available services before designing an AI assignment around it; this is not an unrestricted model-API allowance.
### Redeem a Google award [#redeem-a-google-award]
Google sends staff coupons and a separate student verification link. Follow the [award instructions](https://docs.cloud.google.com/billing/docs/how-to/edu-grants) and check your award's redemption deadline and expiry. Google says education-credit redemption does not require a credit card. A card prompt is a reason to check the [coupon route](https://support.google.com/google-cloud-higher-ed/answer/10322773?hl=en).
After redemption, [check your Cloud project](/docs/implementation/ai-environment-setup/vertex-ai-setup#check-your-project). A successful request shows the connection works; it does not show that credits paid for it. Inspect credit attribution in billing records once usage appears.
## Hand participants one route [#hand-participants-one-route]
Tell the class which route the course uses, in the course brief, before Session 2. Participants who follow a course key should not create a Cloud project; participants with an assigned Cloud project should not buy gateway credits. The [Quick Start](/docs/implementation/foundational-setup/quick-start-checklist) sends them to the right guide once they know which route is theirs.