DeepSeek V4 Pro is now in the Yenisei catalog

We have added DeepSeek V4 Pro, the top model of the DeepSeek lineup, to the catalog. It handles both long reasoning and code well. If the choice is between "smart but slow" and "fast but shallow," V4 Pro is closer to the former.

What's inside

• Context up to a million tokens. Enough to drop in an entire code module with its tests and documentation and ask the model to figure it out, without slicing it into chunks.
• Strong reasoning. Before answering, the model lays the solution out step by step. On math and logic this is noticeable.
• Code. In our internal runs, V4 Pro handles Python, TypeScript, Go, and SQL confidently, is careful with types, and rarely invents library methods that don't exist.

Pricing

Current prices are on the plans and pricing page. With us, 1 credit = 1 ₽, so your budget is counted directly in rubles, with no exchange-rate conversion.

How to call it

The model is available at the usual OpenAI-compatible endpoint:

curl https://api.yenisei.ru/v1/chat/completions \
  -H "Authorization: Bearer $YENISEI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-pro",
    "messages": [
      {"role": "user", "content": "Explain why this SQL does a full scan, and rewrite it."}
    ]
  }'

When to use V4 Pro

Use it for tasks where a mistake is expensive: refactoring while preserving behavior, untangling someone else's legacy code, financial calculations, log analysis with hypotheses. In other words, wherever accuracy matters more than speed.

If you need lots of cheap requests instead, such as classifying support tickets or short chatbot replies, V4 Pro is overkill. For that we have DeepSeek V4 Flash, which gets its own article. And if you want to see the reasoning explicitly, ask the model to write out its steps in the answer or use Kimi K2 Thinking (kimi-k2-thinking).

Russian

We tested Russian separately. V4 Pro writes it cleanly, without calques from English and without sudden switches to Latin script mid-paragraph. Technical texts and code explanations in Russian come out readable and can be handed to users almost without editing.

The model is already in the catalog. You get a key in your account; no separate approval is needed to access the flagship.