As of October 26, 2026, the current Claude models are four generally available models: Opus 5.5 as the default for long work on code and knowledge, Sonnet 5.5 as the faster and cheaper everyday model, Haiku 5.5 as the smallest and cheapest model for high-volume tasks, and Fable 5.1 as the most capable and most expensive model for hard, long-running tasks. On top of that there is Mythos 5.1, which according to Anthropic is the same model as Fable 5.1 with different safeguards and is only granted through access programs for selected US organizations. If you are unsure, Anthropic itself recommends starting with Opus 5.5. Through the API, the models cost between 0.10 dollars (Haiku 5.5) and 10 dollars (Fable 5.1) per million input tokens. On a subscription you do not pay per token; there, your plan decides which models you see in the picker.
I work with Claude every day; my team of fourteen AI employees runs in Claude Code. Further down you will see which model is set for them. What the plans cost is covered in Claude Pricing and Plans Explained; this article is only about the models.
Claude models at a glance: the table
Prices in US dollars according to Anthropic's models overview and pricing page, as of October 26, 2026. For Haiku 5.5 the low price applies to prompts up to 100,000 tokens; above that it costs 0.50 dollars input and 2.50 dollars output per million tokens.
What the four generally available models share: a context window of one million tokens, up to 128,000 output tokens, image input and tool use. Anthropic describes speed relative to each other: Fable 5.1 is the slowest, Haiku 5.5 the fastest. Older models such as Opus 5, Sonnet 5, Opus 4.8 or Haiku 4.5 are still available through the API but are listed as legacy models.
Opus 5.5: the default
Opus 5.5 is the first model of the 5.5 family and has been available since September 22, 2026. Anthropic says two things about it: it performs at the level of Fable 5.1 on most work, and it costs 40 percent less to run than Opus 5. The price per token is 20 percent below Opus 5; according to Anthropic, the rest of the saving comes from the model needing fewer tokens for the same work. That is the vendor's claim, not my own measurement.
In Claude Code, Opus 5.5 is the default model on every plan that includes Claude Code (Pro, Max, Team, Enterprise) and on the API. For urgent work there is a fast mode, according to Anthropic up to 2.5 times faster, priced on the API at double: 8 dollars input and 40 dollars output per million tokens.
In every employee definition of my team it says model: opus. That is not a fixed model but an alias: according to the Claude Code documentation, opus always points to the newest Opus model, currently Opus 5.5 at Anthropic. The upside of such an alias: when a new Opus ships, you do not have to change a single file. The downside: behavior can change with the model switch without you touching anything. If you do not want that, write the full name claude-opus-5-5.
Sonnet 5.5: the faster everyday model
Sonnet 5.5 arrived on September 28, 2026 and costs the same as its predecessor through the API: 2 dollars input and 10 dollars output per million tokens. Content read from the cache now costs half of what it did on Sonnet 5. Anthropic describes its strengths as well-scoped everyday tasks, fixing bugs and creating polished documents, slides and spreadsheets; output is said to be more than 30 percent faster than Sonnet 5.
When a task is clearly defined and you want a quick result, Sonnet 5.5 is the obvious choice. For long, open-ended tasks over many steps, Anthropic points to Opus instead.
Haiku 5.5: small, fast, cheap
Haiku 5.5 was released on October 7, 2026. Anthropic calls it the cheapest, fastest and most capable small model it has released so far. It is built for quick, repetitive work: summaries, database queries, classification, plus time-sensitive things like live customer support. Interesting if you work with several agents: Anthropic explicitly describes Haiku 5.5 as a subagent that supports Opus 5.5 or Sonnet 5.5 on coding work. For complex agentic coding, Anthropic recommends Sonnet 5.5 or Opus 5.5 instead.
Haiku matters mostly on the API, where every token costs money. On a subscription the price difference does not hit you directly; what counts there is how fast you reach your usage limit.
Fable 5.1 and Mythos 5.1: the top tier with access limits
According to Anthropic, Fable 5.1 is the most capable model for code, knowledge work and long problem solving. Through the API it costs 10 dollars input and 50 dollars output per million tokens, and cache reads became 75 percent cheaper. In Claude Code, Fable is not the default on any plan; you select it explicitly with /model fable. Depending on your plan, Fable may then bill to usage credits instead of your included limits, in which case the model picker shows “Requires usage credits”; which plan bills how is in the table further down.
According to Anthropic, Mythos 5.1 is the same model as Fable 5.1, just with different safeguards. Anthropic grants Mythos 5.1 only to vetted organizations: through its life sciences program and, according to Anthropic, “in the near future” through its cybersecurity program. Mythos 5.1 itself is currently available only to a set of US organizations. Mythos does show up indirectly in Anthropic's security product Claude Security: according to claude.com it uses the full model family, including Mythos, without separate access approval, and it is available in public beta for Claude Enterprise. As a model you work with directly, Mythos is practically not on the table for a German or European business.
Which Claude model on which plan
Sources: comparison table on claude.com/pricing and Anthropic's help article on Fable models, as of October 26, 2026. On Max and Team premium seats, Fable is part of the plan: you can use up to 50 percent of your weekly limit on Fable at no extra cost, and Fable uses up the allowance faster than other models. After that you continue with usage credits or switch models. On Pro and Team standard seats, Fable runs on usage credits at API rates from the first message. How Pro and Max limits work is covered in Claude Pro or Max: Limits Explained.
Effort: how thoroughly the model thinks
Besides the model there is a second dial many people overlook: effort. It controls how much Claude thinks before answering. In the Claude apps you choose it in the model menu next to the send button; the levels are Low, Medium, High, Extra high and Max. According to Anthropic's help center, Low and Medium suit routine work, High is the best balance of quality and speed, Extra high is meant for long-running coding and agentic tasks, and Max for the hardest cases. Each model has a recommended level, marked “Default” in the menu.
The catch is stated right there: higher levels take longer and use more tokens, so you reach your usage limit sooner. In Claude Code you set effort with /effort; Opus 5.5, Sonnet 5.5 and Haiku 5.5 start on medium there. On the API the default is medium for Opus 5.5 and Haiku 5.5 and high for Sonnet 5.5 and Fable 5.1.
Which model for what: my recommendation
I would not switch models for every task. The order I recommend follows the logic Anthropic itself states in its models overview:
- Start with Opus 5.5. Anthropic explicitly recommends it when you are unsure, and it is the default in Claude Code anyway.
- If the result falls short, turn up effort first. A higher level is a smaller step than switching models.
- Only then Fable 5.1. Anthropic names Fable for the case where Opus 5.5 at higher effort still falls short. Check beforehand whether your plan bills Fable from your limits or from credits.
- Sonnet 5.5 for clearly defined everyday tasks, when speed matters more to you than the last bit of depth.
- Haiku 5.5 where every token counts: in your own applications on the API, for classification, summaries or as a supporting subagent.
For a solo business owner on a subscription, that usually means: leave Opus 5.5 in place and pick the right effort. How to call models through the API and what that costs is covered in Claude API Key: Setup, Pricing, First Call. How to give individual subagents their own model, for example a cheap researcher next to a thorough reviewer, is in Claude Code Subagents: Best Practices. Which other agent offerings Anthropic has, from the Agent SDK to Managed Agents, is in Claude Agents: SDK, Managed Agents, Duet.
Frequently asked questions
Which Claude model is the best?
According to Anthropic, Fable 5.1 is the most capable generally available model. For most tasks, though, Anthropic recommends Opus 5.5, which by the vendor's account performs at Fable level on most work and is considerably cheaper. So “the best” depends on whether you need speed, low cost or the last bit of depth.
What is the difference between Claude Opus and Claude Sonnet?
Opus 5.5 is meant for long, open-ended tasks over many steps, Sonnet 5.5 for clearly defined everyday work with a quick answer. Through the API, Sonnet 5.5 costs half of Opus 5.5. On a subscription you notice the difference mainly in speed and thoroughness, not on the bill.
Which Claude models are on the free plan?
According to the pricing page, Sonnet and Haiku. Opus and Fable are not included in the free plan. Pro unlocks Opus; Fable runs there through usage credits.
What is Claude Mythos 5.1?
The same model as Fable 5.1, but according to Anthropic with different safeguards. Access is only for vetted organizations through Anthropic's access programs, currently only in the US.
How do I switch models in Claude Code?
With /model in the running session, for example /model sonnet or /model fable, or at launch with claude --model. According to the documentation, the aliases opus, sonnet and haiku point to the newest version at Anthropic; for a fixed model use the full name such as claude-opus-5-5.
Where to go from here
If you are just starting, leave the model as it is and learn to write good instructions first. The starting point is Claude Guide: Getting Started, and for Claude Code Claude Code: A Guide for Beginners. If you want to build models into your own applications, continue with Claude API Key: Setup, Pricing, First Call. And if you want to know how other business owners set up their AI employees and which model runs where: you can ask exactly that in my German-speaking community Claude Practitioners.
