Mistral and variants

[magthe]

They're retiring some models I've been using via ECA and their suggestion is this:

We recommend transitioning to Mistral Medium 3.5 (mistral-medium-3-5 with reasoning_effort set to "high"),

So I modified my config to include

{
  "providers": {
    "mistral": {
      "api": "openai-chat",
      "url": "https://api.mistral.ai/v1",
      "key": "${env:MISTRAL_API_KEY}",
      "models": {
        "mistral-medium-3-5": {
          "variants": {
            "high": {
              "reasoning_effort": "high"
            }
          },
          "extraPayload": {
            "max_completion_tokens": null
          }
        }
      }
    }
  }
}

However when I switch to use the variant I get a lot of extra stuff printed with the response

Without any variant I don't see the same behaviour.

Any ideas of what's causing this? Is it my naive config, ECA itself, or is it upstream returning unexpected data?

[ericdallo]

@magthe it seems that when reasoning is enabled in mistral, they respond in json which is not fully openai chat compatible, I will take a look if we can add a if for mistral in the code

[ericdallo]

@magthe I did a improvement in latest release, it should work nice, LMK