Performance Very Slow compared with Github Copilot Chat

[jsa4000]

Describe the bug

Hi,

I have tested the same prompt using Github Copilot CLI and Github Copilot Chat with the same configuration, same MCPs, tools, agents, prompt, etc.. and while Github Copilot Chat takes 1-2 minutes, Github Copilot CLI takes 12-14 minutes. I have tested multiple times using different OS

I have extracted the logs and asked github copilot to check the timestamps to check the bottlenecks. I have attached the logs and the report generated in the logs.

As a summary, as far I can see so much time it's spend in flushing the events to the session. I don't know if the behavior in Github Copiloit Chat it's the same, what i know it is a lot fastest...

I am using gpt-5-mini model and several custom subagents that are delegated by the primary agent. These makes some calls to mcps and tools

This is the summary part I think it is the bottle neck.

- #1 — 227.882 s gap

  - Prev (line ~8790): 2026-01-14T06:30:59.394Z — "[DEBUG] Flushed 3 events to session ..."
  - Next (line ~8791): 2026-01-14T06:34:47.276Z — "[DEBUG] response (Request-ID ...):"
  - Interpretation: long wait between flushing events and receiving a response from a request to the AI model. Could be a model/API hang, queueing, or network stall.

- #2 — 105.220 s gap

  - Prev: 2026-01-14T06:35:54.504Z — "Flushed 4 events to session ..."
  - Next: 2026-01-14T06:37:39.724Z — "response (Request-ID ...):"
  - Interpretation: another long wait for model/API response.

Do you know what the problem could be? the use of subagents? the model response (even i am not experiencing the same problem in Github Copilot Chat )?

Thakns in advance
Regards

Affected version

0.0.380 Commit: 39ff7b7

Steps to reproduce the behavior

• Run copilot
• Select a custom agent
• Write a prompt

Expected behavior

No response

Additional context

process-1768372062780-8043.log

summary.md

[nellshamrell]

I am also seeing this - it is currently significantly slower (by a factor of 10) than using Copilot with the same model in VSCode.

This is the version I am using:

copilot --version
0.0.384
Commit: 0b21260

[yfoel]

I was still on Version 0.0.36x until yesterday. Then I did an update and it became unusable. It takes ages before it even starts receiving my prompt. It is "Thinking...." for ever before the reasoning process finally starts. When it finally decided to start performance seems to be okay.

copilot --version
0.0.387
Commit: b939fad

EDIT: I just went back to 0.0.366 and the performance massively increased again

[nellshamrell]

For some additional context, I'm seeing this slowdown when running Copilot CLI in WSL (Ubuntu 24.04 LTS)

[jay-m-z]

Also seeing very bad performance compared to VS Code Copilot Chat, multi-minute stalls just like described by OP's summary.md . I use no subagents or custom MCPs. Tested with gpt-5.2 and gpt-5.2-codex on both Ubuntu 24.04 x64 (install script installation) an macOS 26.1 arm64 (brew installation). version 0.0.388

[laz-001]

#960 (comment)

EDIT: I just went back to 0.0.366 and the performance massively increased again

@jsa4000, may i suggest:

I am using gpt-5-mini model

check if this model performs well in 0.0.366

if yes, then edit your title to something like that

[REGRESSION] gpt models performing up to 10 times slower

reproduction:

• execute a prompt with gpt-5-mini in 0.0.366
• execute the same in latest version

[aezur]

I've been using the VSCode extension since beta. I decided to try the CLI this weekend. I ended up here because I was googling trying to find out what I was missing to make the CLI "click" for me. It is significantly slower and an overall worse experience than the IDE extension so i felt like I was just misunderstanding something.

Can I get confirmation this is actually just a regression bug (and worth trying again in a future version) or whether it is a mental model misalignment, and the CLI is doing more behind the scenes that warrants the slowdown?

[laz-001]

it's a regression, potentially only with gpt models. Try opus models or gemini.

[laz-001]

@jsa4000 , this should be closable.

[jsa4000]

@jsa4000 , this should be closable.

sure, close it. thanks

[laz-001]

@jsa4000 , this should be closable.

sure, close it. thanks

i can't. you filed it, should be able to close i think.

[examon]

This appears to be addressed by fixes shipped in v1.0.13 and v1.0.15. v1.0.13 reduced CPU usage during streaming by optimizing spinner rendering and task polling, and v1.0.15 improved end-of-session responsiveness by removing the previous wait before exit. Community reports after these releases indicated the slowdown was no longer reproducing. If you're still seeing significant performance differences compared to VS Code Copilot Chat, please feel free to reopen.