The Mistral lineup in the catalog: Small 3.2, Nemo and Voxtral
Mistral models in Yenisei: Small 3.2 with images and a 256K context, the compact Nemo, and Voxtral for audio. Which one for which task, and what to use instead of the departed Large and Medium.
Yenisei team
The Mistral lineup in the catalog: Small 3.2, Nemo and Voxtral
Mistral is a European developer with its own character: the models are tidy, predictable, strong in European languages and decent in Russian.
Who is who
- Mistral Small 3.2 (
mistral-small-3.2) is the main model of the lineup in the catalog: image input, tool calling, 256K context. Fast and inexpensive. - Mistral Nemo (
mistral-nemo) is the most compact, for embedding into products and for simple tasks where price and speed matter. - Voxtral Small 24B (
voxtral-small) is Mistral's model for audio: it understands speech and documents.
The flagship Mistral Large and Mistral Medium 3.1 are no longer in the Yenisei catalog. When Small 3.2 is not enough, step up to a senior open model from another developer: DeepSeek V4 Pro (deepseek-pro), or, if you need images, Kimi K2.6 (kimi-k2.6) or Llama 4 Maverick (llama-4-maverick).
An invoice from a picture
curl https://api.yenisei.ru/v1/chat/completions \ -H "Authorization: Bearer $YENISEI_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "mistral-small-3.2", "messages": [{ "role": "user", "content": [ {"type": "text", "text": "Extract the data from this invoice as JSON."}, {"type": "image_url", "image_url": {"url": "https://example.com/invoice.jpg"}} ] }] }'Russian language
Mistral holds a solid middle ground in Russian: not the top, but clean, with no obvious blunders or slips into Latin script. For business correspondence, data extraction and summarization it is a perfectly workable tool. We would give literary text to stronger models, but Mistral covers 90% of work tasks.
How to choose
Start with Small 3.2: it covers most scenarios, including images and long input, for little money. Go down to Nemo for high-volume simple requests where the price per million tokens matters. Go up to a senior model from another developer when you hit a quality ceiling. For audio, go to Voxtral.
All three models are in the catalog, accessible with a regular key.