Higher Tier Models in Claude
A beginner-friendly guide to how Claude models are organised into tiers, what sits at the very top of the ladder, how the Mythos tier and Fable fit in, and how to think about choosing the right model for the job.
Imagine a hardware shop with a wall of tools. There is a small pocket knife that is quick and handy, a sturdy everyday toolbox that handles most chores, a heavy-duty workshop machine for serious jobs, and, locked in a back room, a specialist instrument that only a few trained professionals are allowed to use. You would not grab the back-room instrument to open a letter, and you would not use the pocket knife to build a staircase. Choosing the right tool is half the skill.
Claude works in a similar way. Anthropic does not offer just one Claude. It offers a family of models arranged in tiers, from small and fast to large and highly capable, and now to a top tier called Mythos. This guide explains the idea of tiers from scratch, walks up the ladder one step at a time, and helps you understand what the higher tiers are for. Every new word is explained when it first appears, and no technical background is needed.
1What Is a Model Tier?
Before climbing the ladder, let us agree on what a “tier” and a “model” mean.
A model is the trained AI system itself, the thing that reads your message and writes a reply. Think of it as the engine. A tier is a level or grade within a family of models. Just as an airline has economy, premium and first class, a model family can have levels that differ in how capable, how fast and how costly they are to run.
Within a tier there can be several versions, usually shown by a number. A newer version generally brings improvements over an older one at the same tier. So when you see a name such as “Claude Opus 5.5,” you can read it as a tier name (Opus) followed by a version number (5.5). Tier tells you the general class; version tells you the generation.
Think of cars. A brand might sell a small city car, a family sedan and a luxury model. Each has yearly versions. The tier is “city car versus sedan versus luxury.” The version is “this year’s model versus last year’s.” Both pieces of information matter when you shop.
Model
The trained AI system that does the reading and writing. Different models can have very different strengths.
Tier
A level in the family, such as small and quick, balanced, highly capable, or top of the range.
Version
The generation number inside a tier. A higher number generally means a newer, improved release.
Model string
The exact text identifier that developers use to request a particular model when building software on Claude.
Model lineups change over time. Anthropic’s own documentation is always the most reliable place to confirm what exists today, so treat any written guide, including this one, as a snapshot.
2The Ladder at a Glance
Here is the whole family in one picture, from the lowest rung to the highest.
Anthropic’s most recent publicly available models are named Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5 and Claude Fable 5.1. In everyday terms, Haiku is the smaller and quicker option, Sonnet is the balanced everyday option, and Opus is the more powerful option for harder work. Above Opus sits a newer level called the Mythos tier, and Fable belongs to that tier.
flowchart TD
A["Mythos tier: Claude Mythos 5.1 and Claude Fable 5.1"] --> B["Opus tier: Claude Opus 5.5"]
B --> C["Sonnet tier: Claude Sonnet 5.5"]
C --> D["Haiku tier: Claude Haiku 5.5"]
| Tier | Current model | Best described as |
|---|---|---|
| Haiku | Claude Haiku 5.5 | Small, quick and light, suited to simple and high-volume work |
| Sonnet | Claude Sonnet 5.5 | Smart and efficient for everyday use |
| Opus | Claude Opus 5.5 | Higher capability for demanding tasks |
| Mythos tier | Claude Mythos 5.1 and Claude Fable 5.1 | The top of the range, with extra safeguards on the publicly available Fable |
Picture a transport system. Haiku is the nimble scooter for short hops. Sonnet is the reliable car for daily errands. Opus is the powerful truck for heavy loads. The Mythos tier is something like a specialised aircraft: remarkable capability, but handled under stricter rules because of what it can do.
Do not assume a bigger tier is always the best pick. The right choice depends on the job, the budget and how fast you need an answer, as later chapters explain.
3Why Do Tiers Exist at All?
If one model could do everything perfectly, there would be no need for a ladder. The reason is a set of trade-offs.
Running an AI model costs real computing power. Generally speaking, a more capable model is larger or does more work for each answer, which means it can be slower and more expensive to run. A smaller model is cheaper and quicker, but may struggle with the hardest problems. Tiers let people pick a point on that spectrum rather than forcing everyone to pay for the maximum.
Capability
How well the model handles difficult reasoning, long tasks and subtle instructions.
Speed
How quickly the answer arrives. Lighter models often respond faster.
Cost
How much it costs to run. Higher capability usually comes at a higher price.
Safeguards
How much extra protection is needed. The more powerful the model, the more care its release may require.
A restaurant does not use its most senior head chef to butter toast. It keeps that expertise for the complicated dishes and lets quicker hands handle the simple orders. Tiers let you match the effort to the task in the same way.
There is a fourth reason that is easy to overlook, which is responsibility. As models become more capable, they can help with a wider range of tasks, including some where mistakes or misuse could cause real harm. That is why the upper tiers often come with additional safety measures and, in some cases, limited access, as you will see in the later chapters.
4The Lower Rungs: Haiku and Sonnet
To understand the higher tiers, it helps to first know what the everyday tiers are good at.
Haiku: small and swift
Haiku is the lightest tier. It suits tasks where speed and low cost matter most, and where the problem is not very complicated. Examples include sorting messages into categories, short replies, simple summaries and tasks that must run thousands of times a day. Because it is quick, it is also pleasant for live back-and-forth interaction.
Sonnet: the everyday all-rounder
Sonnet sits in the middle. Claude Sonnet 5.5 is described as a smart and efficient model for everyday use, which makes it a sensible default for writing, explaining, coding help, analysis and general questions. For many people and many tasks, it offers a strong balance between quality and cost.
Haiku example: tagging customer emails
A company receives huge numbers of emails and wants each one labelled quickly as billing, delivery or feedback. The task is simple but high in volume, so a light, fast model is a good match.
Sonnet example: drafting and reviewing a report
A manager wants a clear first draft, then help tightening the argument. This needs good writing and reasoning but not the maximum power available, so a balanced model fits well.
Why start low
A sensible habit is to begin with a lighter tier and move up only if the results are not good enough. This saves money and time on the many tasks that never needed more.
If you are unsure which tier to use, test the same task on two neighbouring tiers and compare. The difference, or lack of it, will tell you whether the extra cost is worth it.
5Opus: The Heavy Lifter
Moving up one rung brings us to the tier built for harder and longer work.
Opus is the tier people turn to when a task is demanding: complex reasoning with many steps, long and detailed projects, difficult analysis, or situations where a subtle mistake would be costly. Claude Opus 5.5 is one of the most recent publicly available models and sits just below the Mythos tier.
What does “demanding” look like in real life? It might be reviewing a large body of material and pulling out the contradictions, planning a multi-part project with many dependencies, working through a tricky technical problem, or producing a long, carefully structured document where consistency across many pages matters.
Advantages
- Stronger on complicated, multi-step problems
- Better at keeping track of many details at once
- More reliable on subtle or ambiguous instructions
- Useful when the cost of an error is high
Disadvantages / Trade-offs
- Usually costs more to run than lighter tiers
- May respond more slowly on large tasks
- Overkill for simple or repetitive jobs
If Sonnet is a skilled general practitioner who handles most everyday problems well, Opus is the specialist you see when the case is complicated. You would not book the specialist for a routine check-up, but you would be glad of them when things get difficult.
6The Mythos Tier: Above Opus
Now we reach the top of the ladder, a tier that is treated with noticeably more care.
Anthropic describes a new tier called Mythos that sits above Opus. It represents the most capable class of Claude models. The current generation of Mythos-tier models consists of Claude Mythos 5.1 and Claude Fable 5.1. The first Mythos-class model to appear was called Claude Mythos Preview, and it was not released to the general public.
Why a separate tier rather than just “a bigger Opus”? Because the jump in capability brings a jump in responsibility. A model that is very good at reasoning, programming and scientific work can be enormously helpful, but the same skills could be misused. Placing these models in their own tier signals that they are handled differently, with extra attention to safety and, in some cases, restricted access.
flowchart LR
A["Mythos Preview: limited trusted partners"] --> B["Mythos 5.1: top-tier model"]
A --> C["Fable 5.1: same model with added safeguards"]
B -.-> D["Shared underlying model"]
C -.-> D
Consider a very powerful research laboratory instrument. It is built to do remarkable things, but the lab does not simply leave it on the open shelf. Trained staff use it, rules are in place, and some of its more sensitive functions have extra locks. The Mythos tier is treated in a similar spirit.
The words “higher tier” describe capability and level of care, not a guarantee of being right. Even the top tier can make mistakes, so important results should still be checked.
7Mythos Preview and Project Glasswing
The first Mythos-class model was introduced in a deliberately limited way.
Claude Mythos Preview is not available to the public. Instead, it is used by a small number of trusted organisations as part of an initiative called Project Glasswing. A preview in this sense means an early look at a very capable model, released carefully to selected partners so that it can be studied, tested and used responsibly before any wider decisions are made.
The idea of a limited preview is common whenever a powerful new technology arrives. Rather than switching it on for everyone at once, the developers work with a small group of trusted partners, learn how it behaves in real situations, and build up the safeguards that wider use would need. Anthropic points to its Project Glasswing page for further information.
Develop the model
A new top-tier model is trained and evaluated internally.
Share with trusted organisations
A small number of vetted partners use the preview under agreed conditions.
Learn and strengthen safeguards
Real-world findings feed back into testing, protections and policies.
Decide on wider release
Broader availability, where it happens, follows with measures suited to the risks.
Before a new aircraft carries ordinary passengers, it flies with test pilots and engineers. They push it, find problems and fix them. A limited preview is the equivalent of those test flights.
8Mythos 5.1 and Fable 5.1: One Brain, Different Safeguards
The two current Mythos-tier models are closely related, and understanding the relationship clears up a lot of confusion.
Claude Mythos 5.1 and Claude Fable 5.1 share the same underlying model. In other words, they are built on the same core system. The difference is that Fable adds extra safety measures in three areas: biology, cybersecurity and research and development of large language models themselves.
So Fable is best understood as the version of the top-tier model with additional guardrails in sensitive areas. Fable 5.1 appears among the most recent publicly available models, which makes it the top-tier option that people can reach through Anthropic’s usual products and developer tools.
| Feature | Claude Mythos 5.1 | Claude Fable 5.1 |
|---|---|---|
| Tier | Mythos tier | Mythos tier |
| Underlying model | Shared core model | Same shared core model |
| Extra safety measures for biology, cybersecurity and LLM R&D | Not described as included | Included |
| Listed among most recent public models | No | Yes |
Think of two editions of the same powerful reference library. Both hold the same books. One edition keeps a few sensitive shelves behind a locked glass door with a librarian on duty. That edition is the one placed in the public reading room. The books are the same; the difference is the care taken around certain topics.
When you see two top-tier names, do not imagine two completely different brains. Imagine one strong core, with differences in how it is protected and who can reach it.
9Why Extra Safeguards in These Three Areas?
Biology, cybersecurity and AI research are areas where powerful help can cut both ways.
Biology
Strong scientific help is valuable for medicine and research, but some biological knowledge could be dangerous in the wrong hands, so special care applies.
Cybersecurity
Understanding computer systems helps defenders protect them, yet the same understanding could help attackers, so the line needs careful handling.
LLM R&D
Research into building and improving language models themselves is another sensitive area, because rapid progress here affects how quickly AI capabilities grow.
This is often described as the dual-use problem. A dual-use skill is one that can serve both good and harmful purposes. A chemistry teacher and a wrongdoer might both ask about the same reaction. A security professional and an attacker might both ask how a weakness works. The more capable the model, the more useful it is to both groups, so the developers add safeguards designed to keep the benefits while limiting the risks.
Benefit side
Researchers, students, defenders and engineers receive stronger help with legitimate, constructive work.
Risk side
Certain detailed, high-impact assistance is limited or declined, because the potential harm outweighs the benefit of providing it.
A pharmacy sells many ordinary medicines freely but keeps certain strong ones behind the counter and checks who is asking. The shop is not unhelpful; it is matching the level of care to the level of risk.
Safeguards are not perfect. Sometimes they may be cautious about a harmless request that merely looks similar to a risky one. If that happens, rewording the request with clear, legitimate context can help, and feedback tools in the product let you report cases that seem wrong.
10The June 2026 Timeline
The recent history of the top tier includes an unusual episode that is worth understanding factually.
Claude Fable 5 and Claude Mythos 5 were first released on June 9, 2026. A few days later, on June 12, 2026, Anthropic suspended access to both models in order to comply with export controls from the United States Department of Commerce. Export controls are government rules that restrict the movement of certain technologies across borders or to certain parties. The Department lifted those controls on June 30, 2026, and Anthropic restored access on July 1, 2026. Anthropic published a statement about these events on its website, and that statement is the right place for anyone who wants the official account and further detail.
June 9, 2026 — First release
Claude Fable 5 and Claude Mythos 5 are released.
June 12, 2026 — Access suspended
Anthropic suspends access to both models to comply with U.S. Department of Commerce export controls.
June 30, 2026 — Controls lifted
The Department of Commerce lifts the export controls.
July 1, 2026 — Access restored
Anthropic restores access to the models.
flowchart LR
A["June 9: Release"] --> B["June 12: Access suspended"]
B --> C["June 30: Controls lifted"]
C --> D["July 1: Access restored"]
This episode shows that the most capable AI models are no longer treated as ordinary software. Governments pay attention to them, and rules about who may use them and where can change quickly. For official details, read Anthropic’s own statement rather than relying on summaries.
11How to Choose the Right Tier
Now that the ladder is clear, let us turn it into a simple method for picking a rung.
flowchart TD
A["Describe your task"] --> B{"Is it simple and high volume?"}
B -- "Yes" --> C["Try Haiku"]
B -- "No" --> D{"Is it everyday writing, analysis or coding help?"}
D -- "Yes" --> E["Try Sonnet"]
D -- "No" --> F{"Is it complex, long or high stakes?"}
F -- "Yes" --> G["Try Opus"]
F -- "Needs the very top" --> H["Consider the Mythos tier, such as Fable"]
Define the job clearly
Write down what you want and what a good result looks like. A clear goal makes the choice easier.
Consider the stakes
A casual brainstorm can tolerate small slips. A decision involving money, safety or reputation deserves more care and a stronger model.
Check speed and budget
If you need instant replies or must run the task thousands of times, lighter tiers become attractive.
Test and compare
Run the same prompt on two tiers. If the lighter one is good enough, you have saved effort. If not, step up.
Verify important outputs
Whatever tier you choose, check key facts and conclusions before relying on them.
| Situation | Sensible starting tier | Reason |
|---|---|---|
| Sorting thousands of short messages | Haiku | Speed and low cost matter more than depth |
| Everyday writing and explaining | Sonnet | Strong quality at a balanced cost |
| Long, intricate analysis or planning | Opus | Handles many steps and details better |
| Hardest problems needing top capability | Mythos tier, such as Fable | Highest capability, with extra safeguards in sensitive areas |
12Using Different Tiers in Practice
Choosing is easier when you know where the choice actually appears.
In the Claude app
In the main Claude app on web, desktop and mobile, a person can switch the model they are using, even in the middle of a conversation. This means you could begin a chat with a balanced model and move to a more capable one if the problem turns out to be harder than expected. Which models appear can depend on your plan and settings.
For developers
People building their own software on Claude choose a model by its model string. The recent public models use identifiers such as claude-haiku-5-5, claude-sonnet-5-5, claude-opus-5-5 and claude-fable-5-1. Because identifiers and availability can change, the Anthropic documentation site is the reliable source for what is current.
Claude app
Switch models from within the app, even mid-conversation, depending on what your plan offers.
Developer platform
Select a model string in your software to tell Claude which tier and version to use.
Coding tools
Tools for software engineers let you pick the model that fits the task, such as a quick one for small edits and a stronger one for big refactors.
Partner programmes
Some models, like the Mythos Preview, are reached only through limited programmes with trusted organisations.
A useful pattern for projects is to use a lighter tier for the many small steps and call on a higher tier only for the hardest parts. This can reduce cost without hurting quality where it matters.
13Benefits and Trade-offs of Higher Tiers
The upper rungs offer real advantages, along with costs that deserve honest attention.
Advantages
- Greater ability on the hardest reasoning and technical tasks
- Better handling of long, multi-part work
- Stronger results where errors would be expensive
- Added safeguards on the publicly available top-tier model in sensitive fields
- Potential to help researchers and professionals with serious problems
Disadvantages / Trade-offs
- Typically higher cost per task
- Often slower than lighter tiers
- Access can be limited or change with rules and decisions
- Safeguards may sometimes be cautious about harmless requests
- Not a substitute for checking important results yourself
A professional racing car is astonishing on a track, but it is a poor choice for the school run. It needs special fuel, skilled handling and a lot of money. Higher tiers are the same: wonderful for the right job, wasteful for the wrong one.
14Common Myths About Higher Tiers
A few misunderstandings come up often. Here is how to correct them.
Problem
Believing the highest tier is always the best choice for every task.
Why It’s Harmful
It leads to overspending and slower responses on jobs that a lighter model would handle perfectly well.
Correct Approach
Match the tier to the difficulty and stakes of the task, and step up only when needed.
Problem
Thinking that Mythos 5.1 and Fable 5.1 are two completely unrelated systems.
Why It’s Harmful
It hides the real difference, which is the extra safety measures on Fable, not a separate underlying model.
Correct Approach
Remember that they share the same underlying model, with Fable adding safeguards for biology, cybersecurity and LLM research.
Problem
Assuming that every model Anthropic has built is open to everyone.
Why It’s Harmful
It creates confusion when a model such as Mythos Preview cannot be found, because it is used only by a small number of trusted organisations.
Correct Approach
Check Anthropic’s official pages for what is currently available to you, and treat limited previews as exactly that.
Problem
Trusting any answer without checking, just because it came from a top-tier model.
Why It’s Harmful
Every language model can state wrong things confidently, regardless of tier.
Correct Approach
Verify important facts, especially in areas like health, money, law and safety.
15Frequently Asked Questions
The publicly described ladder runs from Haiku (small and quick) through Sonnet (balanced for everyday use) and Opus (more capable for demanding work) up to the Mythos tier at the top. Fable belongs to the Mythos tier.
As described at the time of writing, they are Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5. Because lineups change, confirm on Anthropic’s documentation before relying on this list.
It is the tier above Opus, representing Anthropic’s most capable class of models. The current generation includes Claude Mythos 5.1 and Claude Fable 5.1.
They share the same underlying model. Fable has additional safety measures in biology, cybersecurity and LLM research and development.
Not as an ordinary member of the public. It is used by a small number of trusted organisations through Project Glasswing, and Anthropic points to its Glasswing page for more information.
They were first released on June 9, 2026. On June 12, Anthropic suspended access to comply with U.S. Department of Commerce export controls. The Department lifted the controls on June 30, and access was restored on July 1, 2026. Anthropic’s published statement has the official account.
In the Claude app, you can switch the model mid-conversation. The models offered may depend on your plan and settings.
No. Choose the tier that fits the task. Lighter tiers are often faster and cheaper, and they are perfectly capable for a large share of everyday work.
Because very capable models can help with both beneficial and harmful tasks, especially in fields like biology and cybersecurity. Safeguards aim to keep the benefits while reducing the risk of serious misuse.
Anthropic’s official website and documentation are the best sources for current models, access and statements. Model lineups and policies can change, so check them directly for anything important.
16Summary and Key Takeaways
Claude is not a single model but a family arranged in tiers. Haiku offers speed and low cost, Sonnet offers a smart everyday balance, Opus brings more power for demanding work, and the Mythos tier sits at the top with the most capable models, Claude Mythos 5.1 and Claude Fable 5.1. These two share one underlying model, with Fable carrying extra safeguards in biology, cybersecurity and LLM research. The earlier Mythos Preview remains limited to a small number of trusted organisations through Project Glasswing. The recent history, including the June 2026 suspension and restoration, shows how seriously the most capable models are treated. The practical lesson is simple: match the tier to the task, begin light, move up when needed, and always check what matters.
Key Takeaways
- Tiers are levels of capability — they let you trade off power, speed and cost instead of paying for the maximum every time.
- The ladder is Haiku, Sonnet, Opus, then Mythos — each step up generally means more capability, usually with more cost and more care.
- Mythos is the top tier — the current generation includes Claude Mythos 5.1 and Claude Fable 5.1.
- Fable and Mythos share a core — Fable adds safety measures for biology, cybersecurity and LLM research and development.
- Mythos Preview is limited — it is used by a small number of trusted organisations under Project Glasswing, not by the general public.
- Availability can change — the June 2026 suspension and July restoration show that access to top models can shift, so rely on official statements.
- Higher is not always better — pick the tier that fits the task, and verify important answers whichever model you use.