All articles
Article
July 1, 20267 min

Claude Opus 5 and Sonnet 5: how they differ and what to use instead in Yenisei

Anthropic's flagship and its balanced smaller sibling. How they differ in practice, and which open models in Yenisei cover the same tasks.

Yenisei team


Claude Opus 5 and Sonnet 5: how they differ and what to use instead in Yenisei

Claude Opus 5 and Claude Sonnet 5 are Anthropic's two top models. Claude is usually chosen for tasks that call for careful step-by-step reasoning.

There are no Claude models in the Yenisei catalog: we work with open-weight models and our own Yenisei lineup. Below is what Opus and Sonnet are good at and what to use instead.

Opus 5, the flagship

  • Reasoning and code in one model. On a hard task it doesn't stop at a surface answer: it works through it step by step and double-checks the result.
  • 1M-token context. A module's codebase, a long correspondence, a hundred-page contract, with no slicing.
  • Images and tools. Suited to agent scenarios and to parsing screenshots and diagrams.

Sonnet 5, the everyday model

The same school, but aimed at a balance of price, speed, and quality. Also 1M context. On most typical tasks the difference from Opus is small, and the answer is cheaper and faster.

What to use in Yenisei

The same "flagship plus everyday model" pair can be assembled from open models:

  • Instead of Opus: DeepSeek V4 Pro (deepseek-pro) or Kimi K3 (kimi-k3): reasoning, tools, a context of about a million tokens; Kimi K3 also takes images as input. Our own alternative is Yenisei Max (yenisei-max).
  • Instead of Sonnet: DeepSeek V4 Flash (deepseek-flash) or GLM 5.3 Flash (glm-5.3-flash): the same scenarios, cheaper and faster, also with a long context.

How to choose

A simple rule:

  1. Draft runs, bulk processing, typical requests: the smaller model.
  2. A task where a mistake is expensive (refactoring while preserving behavior, legal analysis, complex architecture): the larger one.
  3. Not sure? Start with the smaller model and move up to the larger one only where the smaller one starts making mistakes.

Call it from code

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": "Analyze this stack trace and suggest hypotheses about the cause."}
]
}'

Replacing deepseek-pro with deepseek-flash is all it takes to switch to the smaller model. Prices are on the plans and pricing page; payment is in rubles, 1 credit = 1 ₽.