You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I use openrouter, and the cost value shown is always way higher than the actual cost I'm being charged by OpenRouter. I'm not using BYOK for anthropic I'm just using openrouter direct to claude 4 here.
Here's the openrouter data, total comes to 0.064215
Here's what opencode showed for the same session shown above
My experience is quite the opposite; the actual costs within OpenCode are always significantly lower than the deducted totals. I’ve specifically tested this with MiniMax-M2.1 and Kimi-K2.5, and the discrepancy is even more problematic with these models, making it difficult to track actual spend.
Me too,, i am using Mimo-v2-pro on openrouter and the pricing diffrence is huge, opencode says input is 270k open router says i used 3M, keep inmind number of requests match!
@Pxys-io I think when you sup up Cache Read and Input tokens and output tokens and cache write, you would get the 3.77M in open router. For me usually this matches. But price reported by Anthropic SDK in Python vs. open router is very much different
@Pxys-io I think when you sup up Cache Read and Input tokens and output tokens and cache write, you would get the 3.77M in open router. For me usually this matches. But price reported by Anthropic SDK in Python vs. open router is very much different
Yes that was basically it, if cache price wasn't set in models.dev the app assumes is zero, created a PR with fix
Activity
codey-insightdesigns commentedon Sep 8, 2025
I am seeing this issue as well. Costs in opencode show $21, but only about $13 in OpenRouter
Irrelon commentedon Nov 5, 2025
I've just upgraded opencode to the latest version 1.0.23 and can confirm this is still an issue.
awrobel-gd commentedon Jan 26, 2026
I experience the same using openrouter through claude code
radaiko commentedon Jan 31, 2026
My experience is quite the opposite; the actual costs within OpenCode are always significantly lower than the deducted totals. I’ve specifically tested this with MiniMax-M2.1 and Kimi-K2.5, and the discrepancy is even more problematic with these models, making it difficult to track actual spend.
MartinEmrich commentedon Mar 4, 2026
Same with using AWS Bedrock directly. Even in the same session, the $ spent counter sometimes even goes down.
Pxys-io commentedon Apr 3, 2026
Me too,, i am using Mimo-v2-pro on openrouter and the pricing diffrence is huge, opencode says input is 270k open router says i used 3M, keep inmind number of requests match!
awrobel-gd commentedon Apr 3, 2026
@Pxys-io I think when you sup up Cache Read and Input tokens and output tokens and cache write, you would get the 3.77M in open router. For me usually this matches. But price reported by Anthropic SDK in Python vs. open router is very much different
Pxys-io commentedon Apr 3, 2026
Yes that was basically it, if cache price wasn't set in models.dev the app assumes is zero, created a PR with fix
Jaaaky commentedon Apr 3, 2026
I can confirm this too in #20937 Actual is almost double the reported spent credit in opencode :(
Merge pull request anomalyco#454 from different-ai/task/ux-compose-se…