[BUG] Warp systematically ignores selected model -- routes to wrong model/tier across all providers, BYOK Claude key validation broken

[p-kowadkar]

Pre-submit Checks

• I have searched Warp bugs and there are no duplicates
• I have searched Warp known issues page and my issue is not there
• I have an issue with AI and have included the debugging ID (Optional, but helps expedite the AI quality fix). Debugging ID instructions
• I have technical issue and have included the logs (optional, but helps expedite the bug fix). Log instructions

Describe the bug

Warp consistently ignores the model explicitly selected by the user and routes to a different model or tier. This happens regardless of whether a BYOK key is present or not, and affects multiple providers. Adding a Claude BYOK key makes things worse, but the core routing issue exists independently.

1. Model routing is broken across all providers

• Selected Gemini 3 Pro (with Gemini API key, no Claude key) → Warp used Gemini Flash instead
• Selected Gemini 3 Pro (with both Gemini + Claude keys) → Warp routed to Claude Sonnet 4.6, and failed as Claude API key is not validating!
• Selected Claude Sonnet 4.6 Max (with Claude API Key) → Warp internally called Haiku 4.5, and failed!

The pattern suggests Warp is silently downgrading or re-routing to cheaper/smaller model variants regardless of what the user picks.

2. Claude BYOK key validation uses the wrong model
   When a Claude API key is added and Claude Sonnet 4.6 (Max) is selected, the validation error reads: "Failed to authenticate with Anthropic when using claude 4.5 haiku". Haiku was never selected -- Warp appears to be validating against a hardcoded model instead of the selected one.

3. Presence of a failing Claude BYOK key further destabilizes routing
   With a Claude key added (even if valid outside Warp), routing for other providers degrades further -- Gemini 3 Pro failures that don't occur without the Claude key present.

4. Traces confirm multi-model chaos
   Traces during normal usage show multiple unrelated models being invoked in sequence, none consistently matching the selected model.

To reproduce

To reproduce

1. Add a valid Anthropic API key under BYOK settings → observe it fails validation
2. Select Gemini 3 Pro as the agent model → observe Warp routes to a different model -- I've seen it route to Claude Sonnet 4.6 or Gemini 3 flash instead!
3. Switch to Claude Sonnet 4.6 Max with BYOK → observe the error message incorrectly references claude 4.5 haiku

Expected behavior

Expected behavior

• API key validation should correctly validate a working Anthropic API key
• Selecting Gemini 3 Pro should route to Gemini 3 Pro
• Error messages should reference the model actually being called, not a random unrelated model

Screenshots, videos, and logs

This screenshot shows that Claude 4.6 Sonnet (Max) Model was chosen by me, but the validation was done by Haiku 4.5, and that failed too!

Now I tried choosing Gemini 3 Pro (I have API keys set up for Gemini and Claude, and the Gemini API Key validation works, but not Claude! Check this screenshot:

When I remove The Claude API key, that's when it starts working again:

But now, according to the traces:

The fact that I see so many different models, especially when I just try using GPT 5.3 xhigh / Gemini 3 Pro (via my API key)/ Claude 4.6 Sonnet Max (via Warp, since my API key fails all the time), that shows that Warp's model architecture is currently quite unstable!

Operating system (OS)

Windows

Operating system and version

Windows 11

Shell Version

No response

Current Warp version

v0.2026.02.18.08.22.stable_02

Regression

Yes, this bug started recently or with an X Warp version

Recent working Warp date

Warp was working fine with multiple BYOK API keys until mid-Jan to Jan end perhaps.... Haven't had any troubles back then!

Additional context

No response

Does this block you from using Warp daily?

No

Is this an issue only in Warp?

Yes, I confirmed that this only happens in Warp, not other terminals.

Warp Internal (ignore): linear-label:b9d78064-c89e-4973-b153-5178a31ee54e

None

[szgupta]

Hi @p-kowadkar, sorry for the problems here. We try to failover to the next best model when your selected model fails. But agree with you that something seems buggy when your personal API key is invalid -- we shouldn't failover in that case. Do you have the "Warp credit fallback" setting enabled by any chance:

In the meantime, I would make sure that the Anthropic API key that you've added is one that was created on an account that has non-zero credits.

This happens regardless of whether a BYOK key is present or not

Can you send me a repro of this happening when there aren't any API keys registered?

[p-kowadkar]

Hello @szgupta,

Yes, I did have the Warp credit fallback enabled! And I did have both Warp credits and Claude API credits to spare! And I did try it multiple times with different API keys (freshly generated each time from my Anthropic account which had credits), and it always still failed! I just tried adding a new Claude key to Warp, but please pay close attention to the billing part:

It does not say "Billing to my API account" like in case of gemini API key:

A quick suggestion if I could: Next to the API keys, if we had Validate and Save buttons, it would be much more helpful if as a user, I could see if my API key is valid or failing... This feature is especially helpful in case of OpenAI API key, considering OpenAI follows different hyperparameters for reasoning and non-reasoning models (max_completions_tokens vs max_tokens, temperature=1 for reasoning models, from recent models, we could also have reasoning_effort, and so on...

Regarding the problem with incorrect model call, please see the model initialized by me:

And from the traces, gpt-5.3-codex (xhigh) is not even among the top models:

Had I been able to copy the models, or the traces, I'd have pasted it all here, but due to UI limitations, I could only take screenshots! But clearly, the model selected by me is not being even prioritized!