DeepSeek V3.2 in the catalog: an all-purpose chat model with Russian and tools

Update, September 3, 2026: the deepseek-chat name in the catalog now points to DeepSeek V3.2. The model was updated, the name stayed the same.

You don't always need a flagship or a reasoning model. More often, a steady, predictable model that holds a conversation well, can call tools, and doesn't stumble on Russian is enough. That is DeepSeek V3.2, listed in the catalog as deepseek-chat.

Specifications

• Context of about 160K tokens, with room to spare for long conversations and large documents.
• Function calling (tools): the model can decide when to call your function and builds correct arguments.
• High-quality Russian: writes fluently, without anglicisms, with natural syntax.

Why it is a good default

You can safely build a product on V3.2: it handles most requests, is inexpensive, and behaves predictably.

A typical scenario is an assistant with tools:

from openai import OpenAI
client = OpenAI(base_url="https://api.yenisei.ru/v1", api_key="YENISEI_KEY")
tools = [{
    "type": "function",
    "function": {
        "name": "get_balance",
        "description": "Return the user's balance in credits",
        "parameters": {
            "type": "object",
            "properties": {"user_id": {"type": "string"}},
            "required": ["user_id"],
        },
    },
}]
resp = client.chat.completions.create(
    model="deepseek-chat",
    messages=[{"role": "user", "content": "How many credits do I have left?"}],
    tools=tools,
)
print(resp.choices[0].message.tool_calls)

When to look elsewhere

• You need deep reasoning on a hard task: use V4 Pro (deepseek-pro).
• You need many requests, cheaper and faster: V4 Flash.
• You need context close to a million tokens: that is V4 Pro or V4 Flash; both have windows up to a million.

For regular chat with tools, V3.2 is enough, and you don't overpay for what you don't use.