Failed to get response from the AI model; retried 5 times; Last error: Unknown error

[doughgle]

Describe the bug

On 2026-02-23, Copilot CLI repeatedly failed during normal chat turns with automatic retries, burning premium requests without returning usable output.

I observed two failure modes in the same incident window:

• Failed to get response from the AI model; retried 5 times; Last error: Unknown error
• CAPIError: 400 400 Bad Request (with Request-ID), also retried repeatedly
  Example failure:

The issue persisted after exiting Copilot CLI and resuming the session with /resume.
Switching models did not resolve it (Claude Sonnet 4.5, Claude Haiku 4.5, Gemini 3 Pro Preview).
This issue exhausted my premium requests today due to repeated automatic retries and re-sends of context during failure loops.

Affected version

GitHub Copilot CLI 0.0.414

Steps to reproduce the behavior

1. Start Copilot CLI in an existing session with prior context.
2. Run normal prompts, then continue with retry/continue style prompts after a failure.
3. Observe repeated retry loops and eventual query failure.
4. Switch model (Sonnet -> Haiku -> Gemini) and retry.
5. Exit Copilot CLI, then resume session using /resume.
6. Observe failures continue, including repeated 400 Bad Request on later attempts.

Expected behavior

Copilot CLI should either:

• recover and return a response after transient backend errors, or
• fail fast with a clear actionable reason and avoid repeated paid retries for the same invalid/unserviceable request path.

Switching model or resuming session should not keep replaying a failing pattern indefinitely.

Additional context

• OS: Linux
• Shell: bash
• Incident date: 2026-02-23
• Models attempted: Claude Sonnet 4.5, Claude Haiku 4.5, Gemini 3 Pro Preview
• Session behavior: persisted across exit + /resume

Observed errors:

• Failed to get response from the AI model; retried 5 times; Last error: Unknown error
• CAPIError: 400 400 Bad Request

Request-IDs captured from logs:

• 9BE4:71AB1:11F1B88:1569739:699C1D4F
• 9BE4:71AB1:11F45BA:156C643:699C1D53
• 9BE4:71AB1:11F757D:156FB2A:699C1D5C
• 9BE4:71AB1:11FB120:1573D46:699C1D68
• 9BE4:71AB1:12008C3:1579E68:699C1D75
• 9BE4:71AB1:120678E:1580847:699C1D88
• 9BE4:71AB1:1212A42:158E05B:699C1DC5
• 9BE4:71AB1:12153DB:1590E66:699C1DC8
• 9BE4:71AB1:12181A3:1594130:699C1DD2
• 9BE4:71AB1:121B746:1597C52:699C1DDC
• 9BE4:71AB1:121F978:159C676:699C1DE7
• 9BE4:71AB1:122550A:15A2C11:699C1DF6

[dereklegenzoff]

We shipped a fix in version 0.0.416 that should prevent the Response was interrupted due to a server error errors from happening frequently in autopilot mode.

Regarding the 400s, it looks like those errors are predominantly caused by the chrome-devtool-emulate MCP tool. We will look into this issue but in the interim, the best bet might be to disable that MCP tool temporarily.

[dereklegenzoff]

Thanks for providing the request IDs!