Anthropic offers three Claude models for everyday use: Haiku, Sonnet and Opus. Each model is optimised for a different type of task, from quick routine jobs to complex analyses of extensive documents. Which Claude model you use partly determines the quality and speed of your output.
This page gives a technical overview of the regular model lines, including the current versions, the context window and the typical use of each model.
What are Anthropic’s Claude models?
The Claude models come in three model lines, each offering a different balance between speed, capability and cost. Haiku is the fastest and lightest model, Sonnet is the all-rounder for everyday professional use, and Opus is the model for complex and demanding tasks.
The way you use a model affects the quality of your output more than which model you choose. If you want to learn that difference in practice, take a Claude training course that combines model choice, prompts and checkpoints into one way of working.
Since June 2026, there has been a fourth tier above these, which Anthropic calls Mythos. Claude Fable 5.1 is the publicly available variant within that tier and falls outside the everyday choice between Haiku, Sonnet and Opus.
Every model runs on the same technical basis: a large language model trained on large amounts of text and refined using Anthropic’s Constitutional AI method. The difference lies in the scale, the context window and the amount of reasoning the model applies.
What is the difference between the Claude models?
The difference between the Claude models comes down to three things: speed, context window and depth of reasoning. Haiku is fast and light, Sonnet offers a good balance, and Opus is slower but the most capable at complex tasks.
| Model | Best for | Speed | Context window |
|---|---|---|---|
| Haiku 5.5 | Quick tasks, classification, automation | High | 1M tokens |
| Sonnet 5.5 | Everyday professional use, writing, analysis | Medium | 1M tokens |
| Opus 5.5 | Complex analyses, strategic work, heavy coding | Low | 1M tokens |
| Fable 5.1 | Agents that keep working on their own for days | Low | 1M tokens |
In this generation, the context window no longer distinguishes Sonnet from Opus. Both work with one million tokens as standard, with no surcharge for long requests, whereas with earlier versions that was still a reason to switch to Opus.
Pay attention to where you actually get that window. For the claude.ai interface on the paid plans, Anthropic states a context window of up to one million tokens, with the exact window differing per model; through the Claude API, the full million is available on Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5.
For most professionals, the choice comes down to Sonnet, with Opus for specific tasks where depth of reasoning matters more than speed. For AI implementation within a team, consistency is more important than always choosing the newest model: a fixed model per type of task produces more predictable output than letting everyone choose for themselves.
Which Claude model suits your task?
Choose your task and where you use Claude. You will see which model to choose by default and what your plan allows.
What is the Claude Haiku model?
Claude Haiku is the fastest Claude model, designed for tasks that require little reasoning but call for high speed or volume. Examples include classifying texts, extracting structured data, answering simple questions and automating repetitive workflows.
The current version is Haiku 5.5, released on 7 October 2026, with a context window of one million tokens and a maximum output of 128,000 tokens. One million tokens is more than enough for business documents, emails, product descriptions or forms. Haiku responds within seconds and, for prompts up to 100,000 tokens, costs $0.10 per million input tokens and $0.50 per million output tokens, compared with $4 input for Opus 5.5. On claude.ai, you can choose Haiku 5.5 on all plans, including the free plan.
For business integrations, Haiku is available through the Claude API. In practice, Haiku is mainly used in automated workflows where latency and cost matter more than depth of reasoning. For one-off interactive conversations via claude.ai, professionals rarely reach for Haiku, because Sonnet delivers stronger content there for a comparable user experience.
What is the Claude Sonnet model?
Claude Sonnet is the all-rounder within the Claude model family and the most widely used model for everyday professional work. It is suitable for writing and editing reports, analysing data, preparing presentations, drafting customer communication and developing marketing content.
The current version is Sonnet 5.5, released on 28 September 2026 as the successor to Sonnet 5 from June. The model has a context window of one million tokens and an output of up to 128,000 tokens, and on the API it costs $2 per million input tokens and $10 per million output tokens. Anyone can chat with Sonnet 5.5 on Claude.ai, via web, iOS and Android.
According to Anthropic, Sonnet 5.5 runs more than 30 per cent faster than Sonnet 5 and costs up to 30 per cent less for most work. On GDPval-AA, it scores two points below Opus 5.5, and on Terminal-Bench 4.0 it comes out above Opus 5.5 with 70.6 per cent.
In the Claude apps, Sonnet 5.5 runs at medium effort by default, which gives faster answers and lower token usage. If a task needs more reasoning, turn the effort up.
The practical value of Sonnet lies in its balance. The model is fast enough not to interrupt your workflow, but capable enough to deliver output you can use straight away for most business tasks. How well Sonnet works for your tasks depends mainly on the context and instructions you give it.
What is the Claude Opus model?
Claude Opus is the most capable regular Claude model and is intended for tasks that require in-depth analysis, complex reasoning or processing very large amounts of text. The most recent version, Opus 5.5, was released on 22 September 2026 as the successor to Opus 5 from July 2026, and is the first model of the Claude 5.5 family.
Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, twenty per cent less than Opus 5. It has a context window of one million tokens and an output of up to 128,000 tokens, with adaptive reasoning that is always on and adjustable effort levels up to max.
On the subscriptions, Opus 5.5 is available from Pro upwards. For work where latency matters, there is a fast mode that runs up to 2.5 times faster at double the rate.
The model is slower than Sonnet because it uses more compute per answer, but the output is deeper and more accurate on complex, open questions. For strategic analyses, policy notes or tasks where mistakes are costly, that extra processing time is worth it.
Moving to a new Opus version needs attention in practice. With the earlier step from Opus 4.6 to 4.7, I noticed that the answers became longer and less sharp: in my experience, Opus 4.6 was more concrete and kept more distance, whereas 4.7 goes along with the user more readily and phrases things at greater length.
As a result, existing prompts in my workflows had to be retuned to keep the same sharpness in the output. A new model version does not automatically mean existing prompts keep working unchanged, so it is wise to retest your workflows after every model update.
There is one more migration issue that affects your bill. Models from Claude 4.7 onwards use a newer tokenizer that produces about 30 per cent more tokens for the same text, so a request costs more while the price per token stays the same.
What is the Claude Fable 5.1 model?
Since 1 September 2026, Claude Fable 5.1 has been the top model above the Opus family, as the successor to Fable 5, which appeared on 9 June 2026. It is a Mythos-class model, a separate tier that Anthropic positions above Opus for the heaviest reasoning, coding and agentic tasks. Fable 5.1 shares its underlying model with the restricted variant Claude Mythos 5.1, but has extra safeguards around biology, cybersecurity and research into language models, and these allow it to be generally available.
Mythos 5.1 itself is reserved for vetted cyber defenders and life scientists, who get access through the Cyber Verification Program or the Life Sciences Verification Program. Both models cost the same.
On the API, Fable 5.1 costs two and a half times the rate of Opus 5.5, that is, $10 per million input tokens and $50 per million output tokens, with model ID claude-fable-5-1. It supports a context window of one million tokens and an output of up to 128,000 tokens. Those base rates stayed the same in the move from Fable 5 to 5.1. What did change is the price of reading from the cache, which went from one dollar to twenty-five cents per million tokens, or 0.025 times the input rate. Opus 5.5 and Sonnet 5.5 are at one twentieth, Haiku 5.5 and the older generations at one tenth. The change matters most for agents that read the same codebase or set of instructions again and again.
Availability was interrupted in the previous generation. Anthropic suspended access to Fable 5 and Mythos 5 on 12 June 2026 to comply with export controls from the US Department of Commerce. After those measures were lifted, Anthropic restored access to Fable 5 worldwide on 1 July 2026, and access to Mythos 5 only for a group of US organisations after approval by the US authorities. Since September, Mythos 5.1 has been accessible through two access programmes for vetted parties, focused on cybersecurity and life sciences.
On the paid plans, Fable 5.1 counts separately. Pro users pay for usage through usage credits at API rates; on Max and Ultra, the model is available up to fifty per cent of your weekly limits.
For regulated work, there is a limitation you need to know about in advance. Fable 5.1 requires a thirty-day retention period for safety monitoring by default; zero data retention is only possible for eligible customers.
Because of that high rate and the high token usage on long tasks, Fable 5.1 is not a replacement for Sonnet or Opus for everyday work, but an addition for extremely complex tasks.
Which Claude model is in the free version?
In the free version of Claude, you chat with Sonnet 5.5 and can also choose the lighter Haiku 5.5. Opus 5.5 and Fable 5.1 are not available on the free plan.
Sonnet 5.5 appeared on 28 September 2026 as the successor to Sonnet 5. The usage limits apply per rolling five-hour window and depend on how busy the platform is and the length of your conversations.
The paid plans give access to Opus 5.5, higher usage limits and priority for new features. For heavy users, there are Claude Max (€90 a month, five times the usage of Pro) and Ultra (€180 a month, formerly Max 20x, ten times Pro’s weekly usage). For business integrations, all models are available through the API, with token-based billing that differs per model.
Which model works best depends on the task and how you set it up. In the LearnLLM Claude Course, you link the choice of model per task to a way of working with repeatable workflows and checkpoints.
Frequently asked questions about Claude models
If you work with the Claude models, you will probably also have questions about switching models and retirement dates.
How do you switch Claude models during a conversation?
You switch Claude models by clicking the model name next to the send button in the chat window and choosing another model. Under More models, you will find the models that do not appear in the list straight away. According to Anthropic, you can change the model and the effort level at any point in a conversation, and the change applies from Claude’s next answer. In an Enterprise environment, your admin may have switched off certain choices for your role.
Can you still use older Claude models?
You can use older Claude models through the API until their retirement date; after that, a request to such a model returns an error. For publicly released models, Anthropic announces that date at least 60 days in advance, by email and in the documentation. Claude Sonnet 4.5 has been listed as deprecated since 30 September 2026, with 30 November 2026 as the provisional end date and Sonnet 5.5 as its replacement. Amazon Bedrock and Google Cloud (Google managed) follow their own schedule.
Can you also use the Claude models through Amazon Bedrock or Microsoft Foundry?
Yes, besides the Claude API, the Claude models are also available through cloud platforms such as Amazon Bedrock, Claude Models on Google Cloud and Microsoft Foundry. Each model’s page lists the platforms it runs on, and for Opus 5.5 all these platforms are named. On Amazon Bedrock, a model has its own ID, such as anthropic.claude-opus-5-5 instead of claude-opus-5-5. Also check the platform’s own retirement schedule, because it may differ from Anthropic’s.
