One unfinished coding task burned 20% of my monthly quota
OpenCode Go: extremely poor usable coding capacity
I subscribed to OpenCode Go because it is advertised as a low-cost $10/month subscription providing reliable access to popular open coding models.
My practical experience was very different. I gave GLM 5.3 one relatively small implementation plan. The model did not even finish that single task before my 5-hour Go limit was exhausted.
OpenCode's own usage page then showed that this one incomplete task had consumed:
101.3% of my 5-hour allowance, 40.5% of my weekly allowance, and 20.3% of my monthly allowance.
Switching to a cheaper model did not help because the usage limits are shared.
For comparison, I regularly use Codex and Claude for software development and can complete multiple tasks of comparable size per day, as well as much larger implementations, without anything resembling this level of quota consumption.
The $10 headline price therefore gives a misleading impression of value. What matters for a coding subscription is how much engineering work it can actually complete. By that measure, OpenCode Go was extraordinarily expensive: roughly one fifth of the monthly allowance disappeared on one modest task that wasn't even completed.
The published estimated request counts also did not prepare me for this. Agentic coding can apparently turn one user task into enough internal model usage to burn through the quota remarkably quickly.
I cancelled the subscription after this experience.







