max_threads not working since gpt-5.6(-sol)

What version of Codex CLI is running?

codex-cli 0.144.1

What subscription do you have?

X20

Which model were you using?

gpt-5.6-sol

What platform is your computer?

Linux Mint 22.3.0 (Zena)

What terminal emulator and version are you using (if applicable)?

Alacritty 0.17.0

Codex doctor report

_Removed this after solving to save some space_

What issue are you seeing?

Before today I had my max_threads on 50 and I could simply ask Codex to "Spawn one subagent per failing test file" to figure things out. It would spawn, say, 16 subagents, one for each test file and those would then report back.

We are currently limited to 4 threads (so 3 subagents + the main) it seems. I've tried changing stuff around in the config files but to no real avail. Prompts like this still work, but now it gets limited to just 3 in parallel - meaning that getting a report on 16 failing test files kind of takes ages now.

What steps can reproduce the bug?

1. Delete ~/.codex for a fresh start
2. codex --yolo to generate a fresh ~/.codex directory
3. Login to your OpenAI account and close codex.
4. In the newly generated ~/.codex/config.toml add max_threads = 10 under [agents].
5. codex --yolo
6. Ask it to create 10 (no-fork) subagents in parallel that all run sleep 120
7. It will create 3, wait for the 3 to finish, then create another 3, etc etc.

What is the expected behavior?

It creates 9 subagents (10 in total with the main agent) that all sleep.

Additional information

codex doctor --json shows failures because I have codex locally (~/.local/bin) installed instead of globally. This always used to be fine and I don't see how this would affect the max_threads thing.