Claude Fable 5 and models for code and agents: what to use in Yenisei

Claude Fable 5 is Anthropic's model for code and long agent chains, with a 1M-token context.

There are no Claude models in the Yenisei catalog. Below is how Fable differs from general-purpose models and what to use for code and agents with us.

How a "code" model differs from a general-purpose one

A general-purpose flagship is equally good at reasoning, document analysis, and code. A model for development is deliberately shifted toward engineering scenarios:

• it holds long multi-step tasks more carefully: call a tool, look at the result, adjust the plan, continue;
• it loses the thread less often in large edits, including across many files;
• it follows the function-calling protocol more consistently, which matters for agents.

What to use in Yenisei

• Kimi K2.7 Code (kimi-k2.7-code): Moonshot's model tuned for code and development agents: reasoning, tools, images as input.
• Qwen3 Coder Next (qwen3-coder-next): a specialized coding model with tool calling.
• GLM 5.3 (glm-5.3): strong on multi-step agent scenarios, with a window of over a million tokens.
• Yenisei Code (yenisei-code): our coding model with a million-token context.

A mini-agent

from openai import OpenAI
client = OpenAI(base_url="https://api.yenisei.ru/v1", api_key=YENISEI_KEY)
tools = [{
    "type": "function",
    "function": {
        "name": "run_tests",
        "description": "Run the tests and return a report",
        "parameters": {"type": "object", "properties": {}},
    },
}]
resp = client.chat.completions.create(
    model="kimi-k2.7-code",
    messages=[{"role": "user", "content": "Fix the failing test in the billing module and make sure everything is green."}],
    tools=tools,
)
print(resp.choices[0].message)

When to use a code model and when not to

• Use it for: a coding assistant in CI, automated analysis and repair of old code, an agent that works through tools on its own.
• Don't use it for: long texts for people, legal analysis, creative tasks. A general-purpose model such as DeepSeek V4 Pro (deepseek-pro) is a better fit there.

For working right in the terminal there is a ready-made assistant, Yenisei Code. Payment is in rubles, 1 credit = 1 ₽.