Repository navigation
Excessive token consumption since ~Jan 12, 2026 - apparently sending entire project context on every request #8234
Description
Description
Since around January 12, 2026, OpenCode is consuming a ridiculous amount of tokens for simple questions. I'm talking 13,000+ tokens for asking "what's the tech stack?" on a fresh project.
What I Tested
Created a brand new Astro project (default template, minimal files) and asked basic questions in a fresh session:
"What is the tech stack in this project?" → 13,088 tokens
"How do I install packages/dependencies?" → 13,840 tokens
"Why pnpm and not npm?" → 14,161 tokens
Using opencode -c (continuing sessions) is even worse - easily 50k+ tokens per query.
What I've Tried
Started completely fresh sessions with opencode (not -c)
Downgraded between versions
Tested on a minimal project with no history
Nothing helps. It seems like OpenCode is sending the entire project as context every single time, no matter what.
My Setup
OpenCode installed via: npm i -g opencode-ai
Version: 1.1.17 (also tested versions from 1.1.0 to 1.1.16)
Using GitHub Copilot Pro subscription provider
Models: Claude Sonnet 4.5
This is burning through my monthly quota in a single day. Screenshots attached showing the exact consumption.
Is there a way to limit how much context OpenCode sends? Or is this a bug?
Plugins
No response
OpenCode version
1.1.17
Steps to reproduce
No response
Screenshot and/or share link
Operating System
Windows 10
Terminal
Cmder, windows terminal
Activity
github-actions commentedon Jan 13, 2026
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if none of these address your specific case.
rekram1-node commentedon Jan 13, 2026
This isnt a token issue this is jjst the github copilot integration tracking things more accurately, before ut would only yse 1 premium request per session
rekram1-node commentedon Jan 13, 2026
we are working w copilot team to have official integration by tmr in meantime u can downgrade to 1.1.13
adrians5j commentedon Jan 14, 2026
Woah so it's not just me. Good to know.
kennethaasan commentedon Jan 15, 2026
I tried upgrading to v1.1.21 that includes the recent fix, but it's still using a lot of budget vs previously. I'll downgrade to 1.1.13 and stay there until this is a bit more stable. My budget went from 20% something to 70% something in no time...
adrians5j commentedon Jan 15, 2026
I also got wrecked... Like 30% gone in a matter of 10ish mins :|
Did not yet try 1.1.21, but now I think I'll also wait a bit.
P.S. BTW not sure do we maybe need to reconnect with GH Copilot? 🤔
dhruvkej9 commentedon Jan 15, 2026
Yea it is just too much now but 1.1.13 have some bugs so can you tell me the stable version?
1.1.12 & 1.1.13 both have bugs on double escape
hosseinalipour commentedon Jan 29, 2026
I have an idea, what about writing a pause command for the model to pause at the end and program let us respond to the model in that tool? I have done something similar for vscode copilot already.
1.1.13 give me
Error: Token refresh failed: 404