Llama 4 in the catalog: multimodal Scout and Maverick from Meta

The catalog has two fourth-generation Meta models: Llama 4 Scout (llama-4-scout) and Llama 4 Maverick (llama-4-maverick). The main novelty of this generation: both models understand images. The previous-generation model, Llama 3.3 70B, is no longer in the catalog.

Two models, two roles

• Llama 4 Scout is the lightweight one. Fast, economical, understands images, and its context window exceeds a million tokens. For high-volume tasks that need vision at a reasonable price, and for long documents.
• Llama 4 Maverick is the senior one. Stronger on complex tasks with images and long reasoning.

A request with an image

curl https://api.yenisei.ru/v1/chat/completions \
  -H "Authorization: Bearer $YENISEI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama-4-maverick",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Describe what is happening on this sales chart."},
        {"type": "image_url", "image_url": {"url": "https://example.com/chart.png"}}
      ]
    }]
  }'

Russian, honestly

Llama is middling in Russian, and you should know that in advance. Meta's models were trained mostly on English; there is less Russian in the dataset than with the Chinese competitors. On simple tasks everything is fine, but on long or nuanced text, calques and dry phrasing sometimes slip through. For a purely Russian-language product where the text goes to a client, compare with GLM, Kimi or Qwen.

Why Llama then

For the ecosystem. It is the most widespread open standard: thousands of ready-made integrations, prompts, fine-tuned versions and guides have been written for Llama. If you are already building on Llama or want compatibility with open tooling, Scout and Maverick add image understanding without breaking your usual process.

Both are in the catalog, accessible with a regular key.