Claude Sonnet 4.6: what it is and when to pick it over Sonnet 5 and 5.5

Claude Sonnet 4.6 is an earlier version of the Sonnet line. In the Zukko catalog it appears as claude-sonnet-4-6, next to the newer claude-sonnet-5 and claude-sonnet-5-5. If you were looking up what Claude Sonnet 4.6 is, in short: a mid-tier model with a 1M-token window and image input. Below we compare it with the newer Sonnets on price, cache and strength index, and say honestly when it makes sense to stay on it.

We do not retell model features here. We compare only what the catalog table shows as of October 18, 2026, and suggest checking everything else on your own tasks.

Claude Sonnet 4.6: what it is, in one table with 5 and 5.5

Zukko prices, $ per 1M tokens. Cache is billed separately: reads and writes at their own prices. "Official" is the vendor's price list from the same table.

The strength index is the Artificial Analysis Intelligence Index: a general guide, not a guarantee for your task. The table shows three things.

1. Context window and image input are the same for all three.
2. Sonnet 5 and 5.5 cost the same in Zukko, but 5.5 has a higher index (48 versus 45).
3. Sonnet 4.6 costs more than both newer versions on input, output and cache, and has a lower index (42).

There is no Sonnet 4.5 entry in the October 18 table. If you meet that name in old code or a tutorial, check the current list with GET https://api.zukko.pro/v1/models and move to a model from the catalog.

Sonnet 5 and 5.5 compared with 4.6

In the line they are later Sonnet generations. We do not claim they "write better code" or are "smarter" in your scenario, because we do not have your tests. What is objectively known is that the catalog gives them a higher strength index and a lower price. The rest is a matter of testing: run 20-30 of your typical requests through both models and compare.

What the difference costs

Formula: input (millions) × input price + output (millions) × output price.

Example 1. Bulk processing without cache. 1,000 requests, 20,000 input and 500 output tokens each. Input is 20,000 × 1,000 = 20M, output 500 × 1,000 = 0.5M.

• claude-sonnet-4-6: 20 × 0.5 + 0.5 × 2.5 = 10 + 1.25 = $11.25
• claude-sonnet-5 and claude-sonnet-5-5: 20 × 0.4 + 0.5 × 2 = 8 + 1 = $9.00

Sonnet 4.6 costs $2.25 more, which is 25%. At the official prices from the same table it looks like this: for 4.6, 20 × 3 + 0.5 × 15 = $67.50; for 5 and 5.5, 20 × 2 + 0.5 × 10 = $45.00.

Example 2. Agent session with cache. Work through the Anthropic format (/v1/messages, Claude Code, Anthropic SDK), where the gateway sees cache reads and writes. Suppose a 100,000-token context is written to cache once, then 40 requests read it from cache, each adding 2,000 fresh input and 800 output tokens. Totals: write 0.1M, read 100,000 × 40 = 4M, fresh input 2,000 × 40 = 0.08M, output 800 × 40 = 0.032M.

• claude-sonnet-4-6: 0.1 × 0.6 + 4 × 0.063 + 0.08 × 0.5 + 0.032 × 2.5 = 0.06 + 0.252 + 0.04 + 0.08 = $0.432
• claude-sonnet-5-5: 0.1 × 0.5 + 4 × 0.05 + 0.08 × 0.4 + 0.032 × 2 = 0.05 + 0.2 + 0.032 + 0.064 = $0.346

The gap is again about 25%. An important caveat: for Claude the cache discount applies specifically in the Anthropic format. If you call Claude through /v1/chat/completions, all input is billed at the input price. More in the article on cache and formats↗.

When it makes sense to stay on Sonnet 4.6

The honest answer: there are few reasons, and none of them is about price.

• You have a tuned pipeline with tests. Prompts are adjusted to one model's behavior and there is no time to validate a new version. While it works, the move can wait.
• You need reproducibility. You compare results month to month, and a model change would break the series.
• You have not run your tests on the newer versions yet. That is not a reason to stay, it is a reason to run the check.

Keep in mind that vendors retire old models. What the response looks like in that case and how to handle it in code is covered in the article on a retired model↗. Tying production to an old id without a migration plan is risky.

When to move

• You pay more for Sonnet 4.6 and have not measured any gain from it.
• You need a higher strength index at the same or lower price.
• You are starting a new project: there is no reason to begin on an older version.

How to move without breaking production:

1. Put the model name in an environment variable or config.
2. Assemble 20-30 real requests with reference answers or checkable criteria (see the article on a simple eval harness↗).
3. Run the same set through claude-sonnet-4-6 and claude-sonnet-5-5, compare results and cost.
4. Switch a small share of traffic, then the rest.

A switch in Python (OpenAI format):

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.zukko.pro/v1", api_key=os.environ["ZUKKO_KEY"])
model = os.environ.get("LLM_MODEL", "claude-sonnet-5-5")  # was: claude-sonnet-4-6

resp = client.chat.completions.create(
    model=model,
    messages=[{"role": "user", "content": "Briefly explain what idempotency is."}],
    max_tokens=300,
)
print(resp.choices[0].message.content)

More on request parameters: the chat completions docs↗.

FAQ

What is Claude Sonnet 4.6, in a nutshell?

One of the mid-tier versions in the Claude line. In the Zukko catalog: claude-sonnet-4-6, 1M context, image input, $0.5 per 1M input and $2.5 per 1M output tokens, strength index 42.

How does Sonnet 5 differ from Sonnet 5.5?

The price in Zukko is the same ($0.4 and $2 per 1M tokens) and so is the 1M context. 5.5 has a higher strength index: 48 versus 45. Which one handles your task better only a test will show.

Can I call Sonnet 4.6 through the OpenAI format?

Yes. Any model can be called in the OpenAI format (/v1/chat/completions) or the Anthropic format (/v1/messages). Just remember that the cache discount for Claude comes from the Anthropic format.

What happens if the model is retired?

Requests to it start returning an error, and GET /v1/models stops listing it. Keep the model name in config and check the list from time to time.

If you want to compare the versions on your own data, create a key in the dashboard↗: all three models are available on one balance.