Claude Sonnet 5.5 is out: which Claude model to pick for the job
Claude Sonnet 5.5 launched on September 28. There are now several working Claude models in one window. Here is a plain, fact-based guide to which one to use for fast tasks, which for heavy ones, and what it costs.
On September 28, 2026, Anthropic (the company that builds Claude) released a new model, Claude Sonnet 5.5. It arrived six days after Claude Opus 5.5 launched on September 22 and became the default model. You now have several working models in the Claude window, and the question is not which one is "better" but which one to pick for a specific task.
Let us break it down using facts from the official Anthropic model table and the Claude Code changelog, so you do not have to guess.
What exactly launched on September 28?
Claude Sonnet 5.5, a new version of the mid-tier line. Anthropic sums it up in one phrase: the best combination of speed and intelligence. In other words, it answers fast while thinking almost at the level of the heaviest models. In the current lineup its latency is marked "fast," while Opus 5.5 runs at "moderate" speed.
The context window (how much text the model holds in mind during one conversation) for Sonnet 5.5 is 1 million tokens, roughly 555 thousand words. That is the same as Opus 5.5. In a single reply the model can output up to 128 thousand tokens of text.
How is Sonnet different from Opus?
Both models are new, both are version 5.5, both have a 1 million token window. The difference is in purpose.
Anthropic positions Opus 5.5 as the model for long, multi-step work: complex code, large research, tasks that need extended thinking. Sonnet 5.5 is about balance: it is noticeably faster and cheaper, and the quality on everyday tasks is almost as good.
In short: Opus wins on depth, Sonnet wins on speed with nearly the same brain. Anthropic itself advises starting with Opus 5.5 for most workloads and switching only when it falls short.
Which model should you use for which task?
A practical rule for a business owner.
- Fast everyday tasks: a note to a client, a social post, a draft proposal, text edits. Use Sonnet 5.5. It replies quickly and you will barely notice a quality difference.
- Heavy thinking: analyzing a large spreadsheet, strategy, a long document where every detail matters, a contract review. Use Opus 5.5.
- Simple and high-volume: sorting a hundred requests, short replies to repetitive emails. The fastest model is enough here, more on it below.
The point is not to run the heavy model on trivia where it only makes you wait, and not to cut corners on a task that truly needs depth.
What about Haiku?
There is one more model, Claude Haiku 4.5. Anthropic calls it the fastest model with near-frontier intelligence. Its context window is smaller, 200 thousand tokens (about 150 thousand words), but that is plenty for short, high-volume tasks. If you need to run many repetitive operations quickly, Haiku does it fastest and cheapest.
How do you switch models?
In the claude.ai interface the model is chosen from a dropdown above the input box. Open the list, pick Sonnet or Opus, and keep going. You can switch at any moment, even inside the same conversation.
One note: after its release Opus 5.5 became the default on Pro and Team plans, which we covered separately. So if you changed nothing, you already have Opus. Choose Sonnet 5.5 manually from the list when you want faster replies.
What does it cost?
If you use Claude on a claude.ai subscription, you do not pay separately per model: you pick any within your plan. The difference is in usage limits, not in a separate bill.
Per-token prices matter to those working through the API (an interface for connecting programs) or the Claude Code tool. Here Sonnet 5.5 is clearly cheaper: 2 dollars per million input tokens and 10 dollars per million output tokens. For Opus 5.5 it is 4 and 20 dollars, exactly twice as much. So on high-volume API work Sonnet cuts the cost in half.
What should you do right now?
Nothing urgent. Your old conversations and settings are all still there. Just remember the rule: a fast task means Sonnet, a deep one means Opus, a high-volume one means Haiku. Try it once and you will feel the speed difference yourself.
Want to get the most out of Claude for your business? Start with our free materials: the AGINE Academy guides and the first free lesson, where you build your first working artifact in 10 minutes.
AGINE Academy is an independent product, not affiliated with Anthropic. Claude is owned by Anthropic.
Questions
No, not if you use Claude on a claude.ai subscription. You pick the model from the list within your plan. A per-token charge applies only when you work through the API or Claude Code, where Sonnet 5.5 costs 2 dollars per million input tokens and 10 per million output tokens.
As of September 22, 2026, Claude Opus 5.5 is the default on Pro and Team plans. You choose Sonnet 5.5 manually from the model list above the input box.
Use Sonnet 5.5 for fast everyday tasks: notes, posts, drafts, edits. Use Opus 5.5 for deep work: analyzing large data, strategy, long documents where every detail counts. Anthropic advises starting with Opus by default.
Up to 1 million tokens, roughly 555 thousand words. That is the same as Opus 5.5 and five times more than the faster Haiku 4.5, which has a 200 thousand token window.