Mistral Large 4 not working: type validation failed (invalid_union)

[carbogninalberto]

Description

I tried to test via OpenCode Zen Mistral Large 4 and I got the following error:

Type validation failed: Value: {"id":"eb6f5ebd9dbf42b4a66f326daf3823c6","object":"chat.completion.chunk","created":1791357732,"model":"mistral-large-4","choices":[{"index":0,"finish_reason":null,"logprobs":null,"delta":{"content":[{"type":"thinking","thinking":[{"type":"text","text":"The user said \""}],"closed":true}]}}]}.
Error message: [
  {
    "code": "invalid_union",
    "errors": [
      [
        {
          "expected": "string",
          "code": "invalid_type",
          "path": [
            "choices",
            0,
            "delta",
            "content"
          ],
          "message": "Invalid input: expected string, received array"
        }
      ],
      [
        {
          "expected": "object",
          "code": "invalid_type",
          "path": [
            "error"
          ],
          "message": "Invalid input: expected object, received undefined"
        }
      ]
    ],
    "path": [],
    "message": "Invalid input"
  }
]

Plugins

none

OpenCode version

1.18.35

Steps to reproduce

1, open opencode
2. select Mistral Large 4
3. send a hello prompt

Screenshot and/or share link

No response

Operating System

WSL Ubuntu

Terminal

Windows Terminal

[opencode-agent]

Thanks for the report and for including the full error payload, @carbogninalberto!

Version 1.18.35 is OpenCode v1, which now only gets fixes for critical bugs such as data loss, security problems or crashes with no workaround. This one affects a single model while others keep working, so it is unlikely to be fixed in v1.

The cause is that Mistral Large 4 streams its reasoning as a list of typed parts (delta.content: [{"type": "thinking", ...}]) instead of a plain string, and v1's OpenAI-compatible stream parser only accepts strings. For v2 (@opencode/cli, the opencode2 command) there is an open pull request, #53697, that adds support for this format. It isn't merged yet, but it's worth trying v2 once it lands.

If you think this is critical for your setup, or you can still reproduce it on v2, please reply here.

[github-actions]

This issue might be a duplicate of an existing issue. Please check:

• Prompt caching for Mistral is silently disabled: @ai-sdk/mistral 3.x drops promptCacheKey, and 4.x requires a spec-v4 core #50748: Reports Mistral streamed thinking content arrays causing the same expected string, received array validation error in OpenCode.