Already in OpenCode’s provider list.
Connect Standard Compute from inside OpenCode, pick the models you want, and code on a flat monthly budget instead of a per-token meter.
No plugin. No fork. No hand-edited config.
Connect, choose, code.
From nothing to your first request.
Connect
Run /connect in OpenCode, choose Standard Compute, paste your key.
Choose models
Enable the models your key may use. One, or a handful — the router picks between them.
Code
No meter to watch. A fixed monthly budget, and a dashboard showing what is left.
Watch it connect.
Thirty-five seconds, start to first request.
Fast models, not frontier ones.
Measured from OpenCode traffic, ordered by volume — with what that compute costs us today, discounts included.
Per million tokens, at the provider we route to · rates verified 2026-08-29 and they move · you decide which of the 27 models your key may use
The trade-off
A meter never stops you but can bill you anything. A budget is predictable but finite. This is the second kind. If your usage is small and steady, metered may be cheaper — if OpenCode is your all-day driver and the bill is the worry, a fixed price is the point.
What a month costs
Every plan includes more compute than it costs.
Reaching the budget returns HTTP 402 until the period resets. See pricing and fair use.
Reviews
Real customers. Real results.
“We were sending everything through Claude because it was the easiest setup. Once we switched, the simple tasks started going to cheaper models while Claude still handled the harder work. The difference in spend was obvious.”
“My biggest concern was that routing to cheaper models would hurt the output. That hasn't been the experience. The important tasks still get the stronger models, and I don't feel like we've sacrificed quality to save money.”
“I don't want to benchmark a new model every week or maintain routing logic myself. I want to build the product. Standard Compute takes a lot of that decision-making off my plate.”
“I switched away from the $200 GPT plan because I kept running out. I wanted something I could actually rely on for heavier usage without constantly thinking about limits.”
“I expected switching to be a project. It wasn't. We changed the endpoint, kept our existing prompts and workflows, and the agents carried on working. That was probably the biggest surprise.”
Before you connect
Do I have to edit opencode.json?
No. Standard Compute is in OpenCode's provider registry, so /connect lists it — choose it and paste your key. Editing the config by hand still works if you prefer it.
Does tool calling work?
Yes, and it is the normal case rather than an edge case. OpenCode's agent loop runs on tool calls; they pass through unchanged, and most replies we serve to agent clients are tool calls rather than prose.
How much context do I get?
Up to a million tokens on the models above, so a large repository and a long session fit without the agent compacting the conversation early.
Is my code used for training?
No. Standard Compute stores no prompts or responses and does not use your prompts or outputs to train models. See the data privacy page for the full detail.
Which models can I use, and can I pin just one?
27 in the catalog. You choose which ones your key may use — enable a handful and the router picks between them per request, or enable exactly one and everything goes there.
Can I cancel?
Yes, any time, from the billing page. Your plan stays active to the end of the paid period.
Connect it in about a minute.
Grab a key, find Standard Compute in the picker, choose your models.
Comparing APIs for OpenCode · Full setup guide · Data privacy · Pricing