Claude Sonnet: what it is for and which version to pick in 2026

Claude Sonnet is a family of language models from Anthropic that sits well below Opus and Fable in price in the Zukko catalog while still working with a context of up to 1 million tokens. If you ask "what is Claude Sonnet for", the short answer is: it is a model for everyday work with text and code when you need a balance of price and strength. Below we cover which tasks it is typically given, how Sonnet 5.5, 5 and 4.6 differ, and how to choose between them without taking anyone's word for it.

What Claude Sonnet is

Sonnet is one of the Claude model lines released by Anthropic. It is a neural network that accepts text (and images as input) and returns text. You can use it through a chat or through the API, including from terminal agents such as Claude Code. In Zukko you call it by name, for example claude-sonnet-5-5, in either the OpenAI format or the Anthropic format.

What it is usually chosen for

Tasks for which Sonnet is usually chosen, given its balance of price and context:

• Code. Writing and editing functions, explaining someone else's code, tests, diff review. It is often set as the working model in agents and IDE extensions.
• Text. Drafts of letters and articles, rewriting, translation, shortening, summaries of long documents.
• Analysis. Reading contracts, tables, logs, customer requests; extracting fields into JSON.
• Dialogue. Support chatbots where the cost of a reply and its care both matter.
• Agents. Tool-calling loops where the model is called dozens of times per task, so the price of a token matters.
• Images as input. Screenshots, scans, charts: all three versions in the table below accept images.

This describes typical scenarios, not a promise of quality. How well a model handles your particular task is shown only by a test (script below).

Sonnet 5.5, 5 and 4.6: what differs

Prices as of October 18, 2026, $ per 1M tokens:

The strength index here is the Intelligence Index from Artificial Analysis; it is a composite number, not a guarantee of results on your task. Takeaways from the table:

• Sonnet 5.5 and Sonnet 5 cost the same, and 5.5 has the higher index. For new projects claude-sonnet-5-5 is a reasonable starting point.
• Sonnet 4.6 costs more and has a lower index. There is only one reason to stay on it: you already have prompts tuned for it and do not want to touch them until you have tested a replacement.
• Opus 5.5 costs twice as much as Sonnet and is higher on the index. It is justified where Sonnet fails on your examples.

The current list and prices are at GET https://api.zukko.pro/v1/models or on the "Models" page in the dashboard; per-model details are in the models reference↗.

What it costs: an example with the formula

Say you process 1,000 requests, each with 3,000 input tokens and 500 output tokens. That is 3M input tokens and 0.5M output tokens in total.

• Sonnet 5.5: 3 x 0.4 + 0.5 x 2 = 1.20 + 1.00 = $2.20
• Sonnet 4.6: 3 x 0.5 + 0.5 x 2.5 = 1.50 + 1.25 = $2.75
• Opus 5.5: 3 x 0.8 + 0.5 x 4 = 2.40 + 2.00 = $4.40

If a shared part of the request (instructions, reference text) repeats, in the Anthropic format (/v1/messages) reading it from cache is billed separately and cheaper: for Sonnet 5.5 it is 0.05 per million against 0.4 for ordinary input. In the OpenAI format, all Claude input goes at the input price. The details are in the article on prompt caching reads and writes↗.

How to test the choice on your tasks

Take 20-30 real examples from your work and run them through each version. Compare results by eye and count tokens:

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.zukko.pro",  # no /v1 for the Anthropic SDK
    api_key="fetch_your_key",
)

MODELS = ["claude-sonnet-5-5", "claude-sonnet-5", "claude-sonnet-4-6"]
PRICES = {  # $ per 1M tokens: input, output
    "claude-sonnet-5-5": (0.4, 2),
    "claude-sonnet-5": (0.4, 2),
    "claude-sonnet-4-6": (0.5, 2.5),
}
task = "Rewrite briefly and politely: 'We did not manage to process your request, wait.'"

Replace task with a set of your own examples, save the answers and compare. If the cheaper version gives an acceptable result on your set, there is no point paying for a stronger model; if not, move up. A general approach to choosing is described in Sonnet 5 or Opus 5.5↗ and in the piece on the model strength index↗.

How to connect

Create a key in the dashboard↗, top up the balance (SBP, cryptocurrency or a MasterCard) and set the key and address in your client. For the Anthropic SDK and Claude Code the address is https://api.zukko.pro; for OpenAI-compatible clients it is https://api.zukko.pro/v1. Access from Russia goes through the gateway.

FAQ

Is Claude Sonnet the same as Claude?

Claude is the general name of Anthropic's models; Sonnet is one of their lines. The catalog also has Opus and Fable, with different prices and indexes.

Which Sonnet version should I take by default?

As of October 18, 2026, Sonnet 5.5 and Sonnet 5 cost the same and 5.5 has the higher strength index (48 versus 45), so start with 5.5 and test on your examples.

Does Sonnet accept images?

Yes, the catalog table marks image input for all three versions. Check sizes and formats on your own request.

When is Sonnet not enough?

If results on your examples are clearly worse than you need, try claude-opus-5-5. Decide from test results, not from the name.