Installing a tool is not the same thing as getting access to a model.
For example:
copilot, claude, opencode, and llm are toolsIf a tool installs correctly but still cannot answer prompts, missing model access is often the reason.
After working through this page, students should be better able to:
Claude Pro or Claude Max when you want one clear
paid coding-tool path| Route | Good when | Works well with | Main tradeoff |
|---|---|---|---|
| GitHub Copilot | student benefits are available | GitHub Copilot CLI, OpenCode, llm-github-copilot |
access, quotas, and plan details can change |
| Claude subscription | you want one paid coding tool quickly | Claude Code | mainly a Claude Code route, not the general answer for every other tool |
| Direct APIs | you want provider choice or scripts | llm, OpenCode |
you manage billing, keys, and cost yourself |
| OpenCode Zen | you want tested models inside OpenCode | OpenCode | pay-as-you-go and mostly specific to the OpenCode ecosystem |
| Ollama | you want local or offline models | OpenCode, llm-ollama |
requires local setup and enough hardware |
GitHub Copilot is often the easiest route when student benefits are active.
In this course, that route can be reused in several places:
/connectllm through the llm-github-copilot pluginGitHub-backed access often uses browser or device login rather than copying a plain API key. That is normal.
GitHub is still worth trying, but do not assume approval, quotas, or model availability will stay fixed forever.
If you want one polished paid coding agent and do not want to wait on GitHub approval, Claude Code is the clearest path.
Claude Code access requires a plan that includes it, such as Claude Pro,
Claude Max, team or enterprise access, or a Console account. The free
Claude.ai plan does not include Claude Code.
This is best understood as a Claude Code route. It is not the same thing as setting up a general API key for every other tool.
Direct APIs are the flexible route when you want scripting, provider choice, or repeatable automation.
Common examples are:
With direct APIs, you normally:
In llm, this usually means installing the matching plugin and using
llm keys set ... or environment variables.
In OpenCode, this usually means using /connect and then /models.
Important: subscriptions and APIs are different. Paying for a chat app or a coding subscription does not automatically mean you have a separate general-purpose API key.
OpenCode Zen is an optional provider from the OpenCode team.
Its main idea is simple: instead of making you compare many providers yourself, Zen offers a curated list of models that the OpenCode team has tested and verified to work well with OpenCode.
The usual flow is:
/connect/modelsZen is pay-as-you-go. It also has some free models, monthly limits, and auto-reload settings. It can also coexist with your own OpenAI or Anthropic keys, but you do not need that to get started.
Ollama is the main local option worth knowing early.
It lets you run open models on your own machine, which can be useful when you want:
Ollama can be used with OpenCode and with llm through the llm-ollama
plugin.
If you want the deeper local track, read Local models and Ollama.
The tradeoff is that local models depend on your own hardware. This is not the
default route for most students, but it is a good option if you have enough RAM
or GPU and want to learn local workflows. In OpenCode, tool-calling with Ollama
may need a larger num_ctx setting.
Once you know how to reach a model, the next question is which model to actually use for a given task. See Which model should I use?.
Try these before reading the suggested answers:
llm, but it still cannot answer prompts. Which critical
distinction should you check first?Suggested answers:
The table above compares access routes abstractly. In practice, students also want to know which subscriptions feel fair, predictable, and worth the money. The table below collects current pricing, usage limits, and practitioner experience reported in the course discussion channel. Prices, limits, and policies change often, so verify the source links before committing.
| Subscription | Price | Usage / limits | Notes | Sources | Experience and fairness |
|---|---|---|---|---|---|
| OpenAI Codex via ChatGPT Plus | $20/mo | Codex included; lighter use; suitable for select projects throughout the week | Codex is bundled into ChatGPT, not sold as a separate consumer subscription | Codex CLI docs, ChatGPT Pro help | Feels good right now. Ric reports that Codex via ChatGPT Plus finally made the subscription worthwhile and that it feels like working without running out of gas. Currently positive, with explicit expectation that this could tighten later. |
| OpenAI Codex via ChatGPT Pro | $100/mo or $200/mo | Higher Codex usage than Plus; OpenAI currently describes roughly 5x or 20x higher limits than Plus depending on tier | Still bundled into ChatGPT Pro tiers rather than standalone Codex plans | ChatGPT Pro help | No direct first-hand experience reported in the course discussion. Likely fair if you truly need the headroom, but the strongest praise was for Plus-value, not the higher tiers specifically. |
| Claude Pro | $20/mo monthly or $17/mo annual | More usage; includes Claude Code and Claude Cowork | Claude Code Pro is not a separate SKU on the pricing page | Claude Pro pricing | Mixed but still strong. Earlier off-peak promos were popular, and Daniel and Ric both get good work out of Claude, but there is recurring session-limit anxiety and active planning around resets. Productive, but not very predictable. |
| Claude Max 5x | $100/mo | 5x more usage than Pro | Includes Claude Code | Claude Max pricing | Likely better if you value continuity more than price, but the course discussion did not include concrete hands-on Max feedback. The appeal is mainly buying relief from limits. |
| Claude Max 20x | $200/mo | 20x more usage than Pro | Includes Claude Code | Claude Max pricing | Same caveat as Max 5x. Probably attractive only for very heavy use. Claude is admired for output quality, but the pricing and limit model still seems to create more mental overhead than the alternatives. |
| GitHub Copilot Pro+ | $39/mo | 1,500 premium requests/mo; unlimited agent mode and chats with GPT-5 mini; unlimited inline suggestions | Extra premium requests at $0.04/request; usage-based billing changes are coming June 1 | Copilot plans, Copilot billing docs | Sentiment turned negative after the billing change announcement. Daniel and Ric both reacted as if the old deal had been unusually generous and was now getting worse. Predictability seems worse after the multiplier and usage-based billing news. |
| OpenCode Go | $5 first month, then $10/mo | Hard limits: $12 per 5 hours, $30/week, $60/month | Beta | OpenCode Go docs | Positive on workflow. Daniel explicitly praised OpenCode for dynamic context pruning and uses it strategically when Claude is near its limit. The usage model is understandable and the product helps extend productive time rather than creating more anxiety. |
| OpenCode Zen | pay-as-you-go | No fixed monthly cap; per-request billing, free models available, spend limits configurable | More prepaid or PAYG than a normal monthly subscription | OpenCode Zen docs | Probably the most predictable in accounting terms because it is explicit PAYG, but that also means less of the cosy all-you-can-eat feeling. Respectable and transparent, though not discussed emotionally in the same way as Claude or Codex. |
| Google AI Plus | from $7.99/mo | More access; current help pages list up to 30 Pro prompts/day, 90 Thinking/day, 12 Deep Research/day, 50 images/day, 2 videos/day; 128K context | Region-specific pricing | Gemini subscriptions, Gemini usage limits | Mentioned as a plan to include, but no first-hand experience was shared in the course discussion. Fairness unknown. |
| Google AI Pro | $19.99/mo | Higher access; up to 100 Pro/day, 300 Thinking/day, 20 Deep Research/day, 100 images/day, 3 videos/day; 1M context | Region-specific pricing | Gemini subscriptions, Gemini usage limits | Same caveat: included for completeness, but no real usage impressions in the course discussion. Fairness unknown. |
| Google AI Ultra | $249.99/mo | Highest access; up to 500 Pro/day, 1500 Thinking/day, 120 Deep Research/day, 1000 images/day, 5 videos/day, plus Deep Think; 1M context | Region-specific pricing and availability | Gemini subscriptions, Gemini usage limits | No first-hand evidence in the course discussion. Hard to rate fairly. |
A few naming caveats:
From the course discussion:
Plans, model lists, prices, and quotas change often.
Before you rely on a route for a course project or a long workflow, verify the current provider docs and current subscription details.
llm