feat: switch xai provider to responses API

[Jaaneek]

Issue for this PR

Closes #18179 #3890

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Switches xAI provider from completions API to responses API. The reason for this change is to use 'reasoning.encrypted_content' with reasoning models. This should improve performance in long multi turn conversations.

How did you verify your code works?

I did a lot of manual testing also told agent to test the changes e2e. I tried reading files, writing code, opening images etc..

Screenshots / recordings

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

If you do not follow this template your PR will be automatically rejected.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍

[nikolaibogdanov]

@Jaaneek are you sure it's fixed?
I'm now trying it and it still uses chat api for me. The only difference that I use it via ACP.
From opencode logs:

INFO  2026-03-27T10:09:36 +1113ms service=default version=1.3.3 args=["acp"] opencode
...
ERROR 2026-03-27T10:09:46 +88ms service=llm providerID=xai modelID=grok-4-1-fast-non-reasoning sessionID=ses_2d13aa720ffeLji2Rqfqkl6bhc small=false agent=build mode=primary error={"error":{"name":"AI_APICallError","url":"https://XXX/YY/xai/v1/chat/completions","requestBodyValues":{"model":"grok-4-1-fast-non-reasoning","max_completion_tokens":30000,"temperature":0.5,"messages":[{"role":"system","content":"You are a title generator. You output ONLY ..."},{"role":"user","content":"Generate a title for this conversation:\n"},{"role":"user","content":"ping"}],"stream":true,"stream_options":{"include_usage":true}},"statusCode":404,...,"responseBody":"","isRetryable":false}} stream error

It gives 404 just because our proxy doesn't support /chat/completions api.

[Jaaneek]

@nikolaibogdanov will take a look within next 2 hours

[Jaaneek]

@nikolaibogdanov I can't reproduce

cc @rekram1-node can you? I tested ACP via zed with multiple different models from xAI on newest dev branch

[rekram1-node]

Testing rn actually haha

[rekram1-node]

funny seeing the msg come in live

[rekram1-node]

@nikolaibogdanov I cannot reproduce either, the responses api is definitely being used, perhaps you are on an older opencode version?

If u installed using brew make sure u used: brew install anomalyco/tap/opencode

[nikolaibogdanov]

Ok, you are right, it works.
However, we ran into the case, where the custom providers are registered only if the providerKey already registered for this provider. In our case we were setting the keys in the same config file, so the xai provider fell back to the default one and used /chat/completion APIs.
Fixed it with the setting XAI_API_KEY in ENV