Claude Models: Current Lineup, Defaults and Which to Use

The current Claude models are led by Claude Fable 5.2 and restricted-access Claude Mythos 5.2, launched November 9, 2026 at $8 input / $40 output per million tokens. Opus 5.5 remains Anthropic's starting recommendation for most API workloads; Sonnet 5.5 fits everyday app work, and Haiku 5.5 fits high-volume tasks.

Verified against Anthropic's live documentation on November 10, 2026. Claude is Anthropic's family of AI models, available through its apps and through services developers connect to their own software. The current lineup also includes Fable 5.1 and Mythos 5.1 for existing deployments; check the access conditions before choosing a model.

Claude Models Comparison: The Current Lineup

Choose the workload first, then check that your account and platform offer the model. All models below have a 1M-token context window and 128K-token maximum output. These are model specifications, not a subscription's usage allowance. The table combines Anthropic's model overview, its model pages and base API pricing; prices are USD per million tokens, input / output.

Since November 9, 2026, Fable 5.2 is the first pick when you need Anthropic's highest generally available capability for demanding reasoning or a long autonomous assignment. The new pair costs $8 / $40 instead of the 5.1 pair's $10 / $50, a 20% reduction in each base token rate, calculated from those prices. That does not make Fable the economical choice for every task: Opus, Sonnet and Haiku remain cheaper. The November 9 API release notes confirm the launch, limits and migration conditions. Our Claude API pricing guide covers rates in detail.

The Fable 5.2 specifications and the Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 pages list the Claude API, Amazon Bedrock, Google-managed Claude on Google Cloud, Microsoft Foundry, Claude Platform on AWS and Anthropic-managed Claude on Google Cloud. Mythos 5.2 lists the API, Bedrock, Google-managed Claude, Foundry and Anthropic-managed Google Cloud; Mythos 5.1 lists the API, Bedrock, Google-managed Claude and Foundry. App and Claude Code access depend on the plan and organization permissions, separately from those developer-platform listings.

Mythos 5.2 is for vetted organizations. It shares Fable 5.2's specifications and pricing; apply through the verification program covering your use case. An API identifier does not grant access. The Mythos 5.2 model documentation states the restriction; our Claude Mythos explainer covers access in more detail. The Fable 5.1 page labels the earlier model legacy, while the Mythos 5.1 page retains its verification requirement. Neither is an announced retirement.

Fable 5.2 fast mode on the Claude API's Messages API uses speed: "fast", with no beta header, at $16 input / $80 output per million tokens; Anthropic reports up to 2.5x higher output tokens per second.

Before switching to either 5.2 model, check prefix handling: editing system, tools or earlier messages and then returning a later thinking block produces a 400 error, now for every account regardless of age.

Also check retention: both 5.2 models require 30-day data retention and are unavailable under zero data retention unless Anthropic expressly authorizes it.

Which Model Is the Default in Claude Apps and Claude Code?

The app recommendation, the model selected in your account and Claude Code's runtime default are different things. This distinction prevents a common purchasing and rollout mistake: assuming every employee or integration starts on the same model.

Claude Apps: Sonnet for Everyday Work, With Saved-Choice and Admin Overrides

Claude Sonnet 5.5 is the everyday starting recommendation in Claude Academy. This is guidance, not a statement that every app account defaults to Sonnet. The pages read today do not name one universal app default or say Fable 5.2 became it. Use Sonnet as the starting point for daily work. Your next chat can open on a different model because of a remembered choice or organization policy.

Anthropic documents remembered model selections and, for Enterprise, organization or role defaults that can replace them. Administrators can also require each new conversation to start on a chosen model and effort level. Check the name beside the send button to see what your chat will actually use. Default-model behavior.

For a concrete exception, Enterprise's optional efficient mode in the unified Claude experience starts sessions on Sonnet 5.5 at low effort and restricts selection to Sonnet and Haiku. Enterprise configuration.

Claude Code: Opus 5.5, With a Microsoft Foundry Exception

Claude Code, Anthropic's coding assistant, documents these account-type defaults:

• Opus 5.5: Pro, Max, Ultra, Team, Enterprise and the Anthropic API; also Claude Platform on AWS, Amazon Bedrock and Google Cloud's Agent Platform.
• Sonnet 4.5: Microsoft Foundry.

Those are defaults in the absence of a saved selection, organization default or another applicable override. Fable is not the account-type default on any provider. The fable alias now resolves to Fable 5.2 except in Claude apps gateway sessions, where it resolves to Fable 5; select claude-fable-5-2 explicitly there if the gateway serves it. The opus, sonnet and haiku aliases resolve to the latest models on the Anthropic API, but can resolve to older versions elsewhere.

Fable 5.2 requires Claude Code v2.1.321 or later. On Pro and Max, a saved Fable 5.1 user selection moves to the fable alias at the first launch on that version or later; Team and Enterprise wait for organization settings, while Console, third-party and project/local/managed settings keep the explicit 5.1 ID.

Use /model to select a model and /status to inspect the current one. A developer who needs a reproducible starting point can launch with claude --model claude-opus-5-5, subject to provider support. Selection commands.

Claude Models Explained: Capability, Speed and Effort

Choose on the task's failure conditions, then tune how much work the model does. Anthropic names capability, speed, cost and effort as the selection criteria. Its developer guide offers an efficiency-first route starting with Haiku 5.5, and a capability-first route starting with Opus 5.5. If demanding work still fails your evaluations at xhigh or max effort, consider Fable 5.2. Choosing a model.

Anthropic labels Haiku Fastest, Sonnet Fast, Opus Moderate and Fable 5.2 Moderate for comparative latency. These are relative labels, not response-time guarantees. At the base rates above, Opus costs half as much as Fable 5.2 per input or output token, and Sonnet half as much as Opus. Haiku is the lowest-cost option, with a higher rate above its prompt-length threshold. For routine intake classification or extraction, begin with Haiku and move up when it misses a requirement; reserving the larger model for difficult cases is a useful workflow to evaluate.

An evaluation is a repeatable check using work you recognize and an explicit definition of success. For an operator classifying incoming requests, success might mean assigning the correct destination and escalating ambiguous cases. For a founder reviewing a strategic analysis, it might mean preserving the source evidence and flagging conflicting assumptions. These are different acceptance tests even if both tasks arrive as text.

Think of the model as the person you assign and effort as the time you let them spend investigating. Raising effort can improve a difficult answer, but it also increases waiting time and token use. It does not give a model a feature it lacks.

The API effort defaults are high for Fable 5.2 and Sonnet 5.5, and medium for Opus 5.5 and Haiku 5.5. Keep the surface attached to that statement: Anthropic's Sonnet announcement sets Medium as its app and Claude Code default effort, while the Claude Platform uses High. Sonnet 5.5 announcement.

For well-specified Sonnet coding and tool tasks, Anthropic recommends starting at medium, moving higher for harder work, and using xhigh or max where evaluations show a gain. Effort guidance. Record the model and effort together when comparing results; otherwise you cannot tell which change helped.

Best Claude Model for Coding

Start with Opus 5.5 for a complex coding assignment; choose Sonnet 5.5 for well-scoped work where rapid feedback matters. Anthropic positions Opus for extended agentic coding, meaning work where the model plans and uses tools over multiple steps. Sonnet's role combines speed with everyday coding capability. Opus specifications.

For a solo builder fixing a known bug, Sonnet is a sensible starting experiment. For a CTO asking an agent to trace an architectural problem across a repository, start with Opus and judge whether its proposed change survives the project's checks. Move to Fable when the work is unusually extended or your Opus evaluations still reveal consequential gaps. The deeper Claude Opus vs Sonnet comparison explains that boundary.

Best Claude Model for Writing

Sonnet 5.5 is the practical starting point for everyday writing and revision. Anthropic describes it as strong at well-scoped work and producing documents, slides and spreadsheets.

An operator turning an approved brief into an internal announcement should judge factual fidelity, tone and editing effort. A longer response is not automatically a better draft. For simple extraction or a short summary, include Haiku in the evaluation; for writing that depends on difficult analysis, evaluate Opus. Our Claude Haiku vs Sonnet guide develops the everyday-work choice.

Which Claude Model Is Best for Research?

Use Opus for difficult research you will steer interactively; choose Fable 5.2 first when the assignment calls for the highest available capability and an extended finished deliverable. Anthropic's app guide distinguishes interactive analysis from delegated work with fewer check-ins; its current developer selection guide names Fable 5.2 for the highest available capability.

A founder investigating a new market might want to challenge an assumption after every answer. A delegated research assignment might instead specify the source requirements, questions to resolve and final memo. The latter is a better reason to evaluate Fable than simply wanting the highest model name. The earlier Claude Fable 5.1 explainer provides background; use the current 5.2 specifications when designing a new handoff.

What the Context and Output Limits Mean

A context window is working space; maximum output is a ceiling on what the model generates. Tokens are the units a model processes, rather than a fixed number of words. Think of context as the reading table holding your instructions, conversation and source material, with room needed for the response. The output limit caps the generated portion. Context-window documentation.

The standard 1M tokens of context and 128K tokens of maximum output in the table do not mean you can request a million-token report. Thinking also counts toward the API's max_tokens budget, even when its content is not returned.

Anthropic separately documents a Message Batches API beta allowing up to 300k output tokens for Opus 5.5, Sonnet 5.5 and Haiku 5.5 with output-300k-2026-03-24. That is a specific batch feature, not the standard max-output entry or a promise about chat replies. Sonnet model specifications, Haiku model specifications.

For a mid-market operator, the consequence is practical: a model accepting a document collection does not establish that it has extracted every important relationship. Require citations to the source material and review the conclusions that drive a decision. App usage allowances are a separate constraint from these model limits. Usage and length limits.

Where the Models Run

A platform listing establishes a supported route, not your account's permission to use it. Check the deployment and model ID your application actually sends before diagnosing a missing model as a capability problem.

Bedrock Claude Models

Amazon Bedrock lists the current models with identifiers such as anthropic.claude-opus-5-5, including anthropic.claude-mythos-5-2 with verification required. Its access criteria are model-specific. Use the Bedrock ID from Anthropic's table when configuring that route; do not assume the bare Claude API ID is interchangeable in every integration. Claude on Amazon Bedrock.

Vertex AI Claude Models

Anthropic's current documentation labels this route Claude Models on Google Cloud (Google managed) and documents the AnthropicVertex client against an Agent Platform endpoint. Its current model table lists IDs such as claude-opus-5-5, with Mythos access marked limited. Keep the current platform name when recording a deployment, even if your team still calls the integration Vertex AI. Google-managed Claude documentation.

The model overview also distinguishes Anthropic-managed Google Cloud access and Claude Platform on AWS, and lists Microsoft Foundry separately. For a buyer, naming the operator of the service matters because lifecycle schedules can differ.

Older Claude Models: Deprecation and Retirement Dates

Move a deprecated deployment before its retirement date; do not treat every older model as already retired. Anthropic labels supported models Active, models without further updates Legacy, functioning models scheduled for withdrawal Deprecated, and unavailable models Retired. Requests to retired models fail. The following dates are copied from the current deprecations page.

Anthropic recommends claude-sonnet-5-5 as the replacement for the deprecated Sonnet 4.5 API model. That makes it a migration priority for a builder still calling the older ID.

Older Models That Remain Active

Fable 5.1 and Mythos 5.1 remain Active in the deprecations table, with no announced retirement date. The same status table still lists these earlier models as Active, with deprecation shown as N/A. Their retirement field is explicitly “Not sooner than”, followed by these dates:

• Claude Fable 5 and Claude Mythos 5: June 9, 2027. Claude Opus 5: July 24, 2027. Claude Sonnet 5: June 30, 2027.
• Claude Opus 4.8: May 28, 2027. Claude Opus 4.7: April 16, 2027. Claude Opus 4.6: February 5, 2027. Claude Sonnet 4.6: February 17, 2027.
• Claude Opus 4.5: November 24, 2026. Claude Haiku 4.5: October 15, 2026.

A lower-bound date is not an announced shutdown. In particular, Haiku 4.5 remains Active in the API table even though its listed earliest-retirement bound has passed.

App removal is another separate event: Anthropic says Haiku 4.5 left the Free, Pro and Max app model selectors on October 29, 2026, replaced by Haiku 5.5. That change does not affect Claude Code, Team, Enterprise or the API. App-selector notice.

What's Overhyped About Choosing a Model

The biggest model name does not settle the deployment decision. A founder needs an answer that survives review; an operator needs predictable handling of routine cases; a builder needs an integration that keeps working.

The specification table also does not make an upgrade a one-line substitution. Opus 5.5 cannot disable thinking. Sonnet 5.5 uses between_tools to turn off up-front thinking instead of accepting disabled. Both reject forced tool-use settings, and both change how text between tool calls reaches your application. Opus migration changes, Sonnet migration changes.

For a production agent, the relevant question is whether its tools, progress messages and failure handling still work. A higher model tier cannot repair a request your integration sends incorrectly. Keep a working deployment while you validate its replacement; accelerate that work when its current model has a retirement deadline.

Your Monday Move

Record the model, provider and effort used by a workflow that matters, then evaluate a deliberate alternative. Make the decision from completed work.

1. 1

   Inspect the current selection

   In Claude, read the model beside the send button. In Claude Code, check /status. In an API application, inspect its configured model ID and provider.

2. 2

   Define a passing result

   Choose a representative task with an answer you can assess. Include an ambiguous case and specify what should trigger human review. A useful acceptance rule beats a preference for longer answers.

3. 3

   Change one variable

   Compare a model at its starting effort, then tune effort if needed. Keep the prompts and source material consistent. For an existing agent, verify its tool calls and progress reporting before rollout.

4. 4

   Act on the result

   Adopt when the alternative clears your acceptance criteria. Wait when the current active model meets the need and the replacement has not been validated. Prioritize migration when the current ID is deprecated.

Frequently Asked Questions

What are the different AI models?

Within Claude, the public choices are the Haiku, Sonnet, Opus and Fable families. Each serves a different balance of responsiveness and capability. Mythos has restricted organizational access; the current versions and IDs are in the table above.

Is Claude better than GPT-4?

A brand-wide verdict will not tell you whether a particular Claude model meets your requirements. Use the actual task, required features and acceptable errors to evaluate a candidate. This guide compares choices within Claude.

What kind of models does Claude have?

The current models accept text and images and produce text, with vision, multilingual capabilities and tool use. The family name identifies the model line; the API ID identifies the model your integration requests.

How much are Claude models?

See our Claude API pricing guide for the current rates and usage examples.

Is there any model better than Claude?

There is no useful universal answer without a defined job and success criteria. For this decision, measure whether the chosen model handles your normal work and difficult exceptions, and whether its response time fits the workflow.

Can I use Claude models for free?

Yes. Anthropic's learning guide says Claude Free includes Haiku and Sonnet; Opus and Fable require access through paid plans. Availability in an app does not establish access to every developer platform.

How long can we use Claude for free?

There is no fixed conversation duration to promise. Usage depends on factors such as message and file size, conversation length, the model and features used. Read the account's limit notice rather than planning around an assumed number of messages.

What models come with Claude Free?

Haiku and Sonnet are the included families in Anthropic's current learning guide. Their current versions are Haiku 5.5 and Sonnet 5.5; use the model menu to see the choices available to your account. Model menu.

Can I run Claude models locally for free?

The documented routes for these models are hosted apps and services. Running a client such as Claude Code on your computer is different from running the model itself on local hardware. The model pages above do not offer downloadable model weights.

Join the newsletter for practical guides to choosing and using AI at work.