This lesson's interactive features are locked, please to keep using them

Cost Effectiveness

With each new prompt, you can quickly switch models with /models and variants with ctrl+t. Those are your two main levers for controlling cost.

More complex task? Smarter model, higher reasoning.
Less complex task? Cheaper model, lower reasoning.

You can waste a lot of time micromanaging this on every prompt. I just use a sane default for most work.

Monitoring Costs

As you're watching OpenCode do its thing, keep an eye on the right sidebar's "Context" section. You'll see something like this:

Context
147,529 tokens
37% used
$5.56 spent

This tells me that my current session, OpenCode's term for a conversation, has an estimated token (API) cost of $5.56 so far, which is not necessarily what you pay under a subscription due to the subsidy.

I/O and Compute Aren't Expensive

Many developers new to AI assume time = money in coding sessions. Not really. OpenCode can spend a lot of time on a task without creating a huge inference bill. You pay for tokens, not time.

For example, these things can take a while without adding much inference cost:

Running computationally heavy test suites
Executing SQL queries that take a long time to return
Making many long-running network calls