Podcast generation failure (Mistral setup)

[DenoBeno]

I asked about mistral-only setup earlier and managed to put together a setup that uses a combination of self-hosted models (for chat, embedding..) and mistral TTS and STT models.

TTS: voxtral-mini-tts-2603
STT: voxtral-mini-2602

Short problem description: Podcast generation fails with "Mistral API error: HTTP 404"

Detailed description

Adding sources and "chatting with documents" in notebooks works. I've encountered some issues with chat complaining about data being too long for the context, but not related to the topic at hand.

So, I set up a notebook with a couple of documents and used it to generate a "lesson" - up to and including complete booklet design to accompany the lesson(*) . Next thing I wanted to try is to generate a podcast based on this notebook. I set up the speaker and the episode profiles as shown below and tried to generate a podcast.

Speaker profile: solo expert, using voxtral-mini-tts-2603 and "casual_male" Voice ID. Everything else is left at defaults.
Episode profile: solo expert. I tried different outline and transcript models, including mistral-medium, mistral large-2512, and self hosted gemma-4-31B-it. Everything else is left at defaults.

Whatever I do, generation always ends up with "Mistral API error: HTTP 404". What surprises me the most is that nothing is shown in "outline" or "transcript". I kind of anticipated that outline is produced first, then transcript and then TTS model is called and this is where the failure occurs. But no, there is no outline and no transcript.

To be clear, I tested the model setup and this is what I get every time:

(*) Since open notebook can't produce anything visual, I set up a google gemini notebook with same materials and asked it to produce some slides based on this booklet instructions text. That came out pretty good.