Repository navigation
Inference speed is slower than expected with GPT-4.1 in Opencode #872
Description
I have observed that model inference speed in Opencode (using Copilot and GPT-4.1) is noticeably slower than expected. When using Copilot with Claude via my proxy, the inference speed is much higher. Since GPT-4.1 is known for fast performance, something may be holding down the models in Opencode. Please investigate possible bottlenecks, such as API throttling, network delays, or code inefficiencies.
Reacted by Antoine Tavernier
Metadata
Metadata
Assignees
Labels
No labels
Activity
drewbitt commentedon Jul 11, 2025
What is the tps you are seeing?
netanelbi commentedon Jul 11, 2025
running:
opencode --model github-copilot/gpt-4.1 run "write 1k drama story about apple"
took ~17-22 seconds.
on claude it took 14s. thats fine i guess
but openning opencode and doing the same request using the interface took 43 seconds
Uranium2 commentedon Jul 12, 2025
Hello,
Just tried opencode for few days, and it does feel like Copilot GPT-4.1 is slow. I don't have other model to compare on Opencode. But when using Copilot GPT-4.1 directly on Vscode, it does seem faster.
Also, it feels like my CPU fan is bursting when asking a task to Opencode. And i think that the files are modified fast, but in the terminal, the display/writing of "code" that the model will create is slow. Most of the time, the file is already edited and ready, but the console is still writing code (and most of the time the code written doesn't seem to be 100% what he updated on the file)
Yorizel commentedon Jul 16, 2025
i also facing this
Uranium2 commentedon Jul 17, 2025
Pretty sure it's the same issue here
#811
adamdotdevin commentedon Jul 17, 2025
yeah, this is an issue with rendering performance on some systems; can you all share what OS/etc you're using in #811, going to close this as duplicate
docs: split docs entrypoints by audience (anomalyco#872)