Bug Description
When running grok-cli against terminal-bench, the model takes an unusually long time to start completing tasks and fails even the basic hello-world sanity check.
Steps to reproduce
- Install and set up terminal-bench as per the README.
- run command : tb run -a grok-cli -m grok-4-latest -d terminal-bench-core==0.1.1
- Observe that progress is extremely slow — around 6% in 1 hour — and the hello-world task does not pass.
Expected Behavior:
The model should quickly start producing completions for tasks.
The hello-world task should pass as a basic sanity check.
Actual Behavior:
The run hangs for long periods between actions.
Progress is very slow (6% in ~1hr).
This worked before you guys fixed the previous hanging issue I came with.