AI provider
Anthropic Provider
Anthropic is an AI safety and research company that develops Claude, a family of AI models built for reasoning, coding, knowledge work, multimodal applications, and agentic workflows.
- Models
- 13
- active Anthropic entries
- Largest context
- 1M
- tokens in current catalog
- Current family
- Claude 5.5
- frontier through efficient
- Access
- Your key
- connect your Anthropic API key
01 / Overview
About Anthropic
Anthropic is an AI safety and research company that develops Claude, a family of AI models designed for demanding reasoning, coding, knowledge work, multimodal applications, and agentic workflows.
Anthropic describes its goal as building AI systems that are reliable, interpretable, and steerable. The company combines frontier model development with research into alignment, interpretability, model behavior, societal impacts, and safeguards for increasingly capable AI systems.
For developers, Anthropic is primarily represented by the Claude model family and the Claude API. Claude models are used for software engineering, document analysis, research, professional knowledge work, customer-facing assistants, tool use, automation, and long-running agents.
Anthropic structures its model lineup around different workload requirements rather than a single model for every application. Higher-capability models are intended for difficult reasoning and complex work, while faster and less expensive models target everyday or high-volume tasks. That makes model selection an engineering decision involving quality, speed, context requirements, reliability, and cost.
An AI company built around safety research
Anthropic was founded in 2021 and is organized as a Public Benefit Corporation. The company states that its purpose is the responsible development and maintenance of advanced AI for the long-term benefit of humanity.
Safety is closely integrated with Anthropic's model development. Its research includes alignment, interpretability, frontier red teaming, economic impacts, and the behavior of advanced models. Anthropic also publishes system cards and transparency material describing model capabilities, evaluations, and deployment safeguards.
This research focus does not make Claude only a research product. Anthropic deploys Claude as a commercial model family for developers, organizations, and end users. The same models can be accessed through Anthropic's own platform and, depending on the model, through major cloud platforms.
For an application developer, the practical result is a catalog with several capability tiers and large context windows. The useful way to choose between them is empirical: identify realistic candidates, test them on representative tasks, and compare the quality and cost of successful outputs.
- Provider
- Anthropic
- Founded
- 2021
- Primary family
- Claude
- Current family
- Claude 5.5
- Company type
- Public Benefit Corporation
- Focus
- Reliable, interpretable, steerable AI
02 / Evolution
How Anthropic Models Evolved
Claude has evolved from a general-purpose assistant into a broader model family covering frontier reasoning, coding, long-context work, fast everyday tasks, and long-running agents.
Claude 1 and Claude 2: establishing the Claude platform
The early Claude generations established Anthropic's approach to general-purpose language models and developer access. Claude 2, released in 2023, expanded the platform's usefulness for long documents, analysis, writing, and professional workflows.
These early releases helped establish characteristics that remained important in later generations: strong instruction following, long-context work, and an emphasis on predictable model behavior.
Claude 3: Opus, Sonnet, and Haiku
In 2024, Anthropic introduced the Claude 3 family with three clearly differentiated model tiers: Opus, Sonnet, and Haiku.
Opus represented the highest-capability tier, Sonnet targeted a balance between intelligence and speed, and Haiku focused on fast and economical execution. Claude 3 also expanded multimodal capabilities, allowing the models to work with image input as well as text.
For application developers, the family structure introduced a useful pattern: choose a tier according to the difficulty and economics of the workload rather than defaulting to the largest model.
Claude 3.5 and 3.7: stronger coding and agentic behavior
Later Claude 3.x releases improved coding, tool use, reasoning, and computer interaction. These models increasingly became relevant not only for conversational assistants but also for software-development agents and workflows that need to perform several actions before completing a task.
The distinction between a good single response and a good agent became more important. Agentic workloads require a model to maintain task state, inspect results, select tools, recover from errors, and continue toward a goal.
Claude 4 and 4.x: long-running professional workflows
The Claude 4 generation pushed further into coding, professional work, and agents. Anthropic released multiple Opus and Sonnet iterations, with later models providing larger context windows and stronger performance on difficult, multi-step tasks.
Claude Opus 4.x models targeted high-capability work, while Sonnet models continued to occupy a more balanced position for production systems that needed strong performance at higher volume. Haiku remained the efficient tier for workloads where latency and cost mattered more than maximum capability.
Claude 5: broader specialization
The Claude 5 generation expanded the family further. Anthropic introduced Claude Fable 5 for advanced coding, knowledge work, and long-running problem solving, alongside Claude Opus 5 and Claude Sonnet 5.
Fable added another high-capability option for ambitious technical and professional workflows. Opus continued to target difficult work requiring strong reasoning and judgment, while Sonnet focused on capable everyday and agentic workloads at a lower cost.
Claude 5.5 and Fable 5.1: current generation
In September 2026, Anthropic introduced Claude Fable 5.1, Claude Opus 5.5, and Claude Sonnet 5.5. Anthropic positions Fable 5.1 for advanced coding, knowledge work, and long-running problem solving; Opus 5.5 for complex work requiring careful judgment; and Sonnet 5.5 as a faster, lower-cost option for well-scoped everyday work.
Anthropic introduced Claude Haiku 5.5 in October 2026 as its fastest and least expensive small model for high-volume and cost-sensitive workloads.
Not every model released by Anthropic is necessarily available in EidoStack at the same time. The table on this page is generated from EidoStack's current model registry and therefore represents the models that EidoStack actually supports.
- Claude 1 / 2
General-purpose Claude
Long-document analysis, writing, instruction following, and developer API use.
- Claude 3
Opus, Sonnet, and Haiku
A clear model hierarchy covering maximum capability, balanced performance, and efficient execution.
- Claude 3.x
Coding and agents
Stronger coding, tool use, computer interaction, and multi-step workflows.
- Claude 4.x
Long-running work
More capable coding, professional work, agents, and larger working contexts.
- Claude 5
Broader specialization
Fable, Opus, and Sonnet cover different kinds of advanced professional and agentic work.
- Claude 5.5
Current generation
Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 extend the range from frontier work to high-volume execution.
03 / Models
Anthropic Models Available in EidoStack
EidoStack supports Anthropic models across Fable, Opus, Sonnet, and Haiku families. Use the table below to compare the Claude models currently available in EidoStack by pricing, context window, and output limits.
13 Anthropic models
| Model | Provider | Input price / 1M | Output price / 1M | Context window | Max output |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Anthropic | $10.00 | $50.00 | 1,000,000 tokens | 128,000 |
| Claude Fable 5 | Anthropic | $10.00 | $50.00 | 1,000,000 tokens | 128,000 |
| Claude Opus 5.5 | Anthropic | $4.00 | $20.00 | 1,000,000 tokens | 128,000 |
| Claude Opus 5 | Anthropic | $5.00 | $25.00 | 1,000,000 tokens | 128,000 |
| Claude Opus 4.8 | Anthropic | $5.00 | $25.00 | 1,000,000 tokens | 128,000 |
| Claude Opus 4.7 | Anthropic | $5.00 | $25.00 | 1,000,000 tokens | 128,000 |
| Claude Opus 4.6 | Anthropic | $5.00 | $25.00 | 1,000,000 tokens | 128,000 |
| Claude Opus 4.5 | Anthropic | $5.00 | $25.00 | 200,000 tokens | 64,000 |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | 1,000,000 tokens | 128,000 |
| Claude Sonnet 5 | Anthropic | $2.00 | $10.00 | 1,000,000 tokens | 128,000 |
| Claude Sonnet 4.6 | Anthropic | $3.00 | $15.00 | 1,000,000 tokens | 128,000 |
| Claude Sonnet 4.5 | Anthropic | $3.00 | $15.00 | 200,000 tokens | 64,000 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | 200,000 tokens | 64,000 |
The models table must be generated from
AI_MODELSby filtering for the provider identified byproviderId. Model names, prices, context windows, max output values, and links must not be duplicated in this Markdown file.
04 / Choosing
How to Choose an Anthropic Model
The right Claude model depends on how difficult the task is, how quickly it needs to run, how much context it requires, and how much each successful workflow can cost.
A useful starting point is to identify the type of work your application performs and then shortlist two or three Claude models that fit that workload.
Start with the difficulty of the work
For complex coding, difficult reasoning, long-running analysis, and high-value professional workflows, begin with Anthropic's higher-capability models.
Claude Fable 5.1 is positioned for advanced coding, knowledge work, research-oriented tasks, and long-running problem solving. Claude Opus 5.5 targets complex work where careful judgment and strong performance are more important than minimizing cost.
These models make the most sense when the value of getting the answer right is high enough to justify a more capable model.
Use Sonnet for balanced production workloads
The Sonnet tier is designed around a different trade-off. It aims to preserve strong capabilities while running faster and at a lower cost than the highest-end Claude models.
That makes Sonnet models natural candidates for coding assistants, production agents, document workflows, business applications, and other systems that need strong model behavior at meaningful request volume.
A common evaluation strategy is to test a Sonnet model against a higher-capability model on the same examples. If Sonnet consistently meets the required quality threshold, the lower cost can make it a better production choice.
Use Haiku for speed and high-volume tasks
Haiku is Anthropic's efficient model tier. It is intended for workloads where low latency, throughput, and cost matter more than maximum reasoning capability.
Typical candidates include classification, extraction, short transformations, summaries, routing, database-related tasks, simple support interactions, and lightweight subagent work.
The current EidoStack table shows which Haiku models are actually available in the product. Anthropic may release newer models before they are added to EidoStack, so provider availability and EidoStack support should be treated as separate questions.
Evaluate Fable separately from the traditional tiers
Fable is designed for some of Anthropic's most ambitious coding and knowledge-work workloads. It should not be treated simply as a renamed Opus tier.
For applications considering Fable, test the tasks that justify its capabilities: difficult software engineering, long-horizon problem solving, complex knowledge work, and workflows where the model must remain productive across many steps.
If a less expensive Opus or Sonnet model reaches the same quality threshold on your workload, that model may still be the better production choice.
Consider the whole agent, not only the first response
Claude is frequently used in agentic applications, and agent quality cannot be measured by evaluating one answer in isolation.
A useful agent evaluation should test whether the model can:
- understand the overall objective;
- select appropriate tools;
- use tool results correctly;
- maintain relevant state across steps;
- recover when an action fails;
- avoid unnecessary actions;
- produce a correct final result.
A model that is slightly more expensive per request can be cheaper per completed task if it requires fewer retries or fewer corrective steps.
Match context size to the real workload
Large context windows are valuable for codebases, long documents, accumulated conversation history, research material, and retrieval-augmented applications.
But the maximum supported context window is only a capacity limit. It does not mean every application should fill it.
Sending unnecessary context can increase cost and make relevant information harder to distinguish from noise. Measure the context your real application sends and test whether additional context actually improves the result.
Plan for model lifecycle changes
Anthropic regularly releases new Claude generations and updates existing tiers. For production systems, model migration should therefore be part of the architecture rather than an emergency task.
Keep a representative evaluation set and run it against a replacement model before changing production model IDs. Compare quality, latency, token usage, and cost, and look specifically for regressions on difficult examples.
- 01
Task difficulty
Use higher-capability models when reasoning quality or task value justifies the additional cost.
- 02
Workload type
Separate simple generation, coding, knowledge work, and long-running agentic workflows.
- 03
Latency
For interactive or high-volume workloads, response speed can be as important as maximum capability.
- 04
Context
Measure how much working context your application actually needs instead of selecting by maximum size alone.
- 05
Cost
Compare the cost of a successfully completed task, including retries and multi-step execution.
- 06
Evaluation
Run the same representative prompts across shortlisted Claude models before choosing one for production.
05 / Pricing
Anthropic Pricing and Context Windows
When comparing Claude models, focus first on how much your requests cost and how much information each model can process at once.
How Claude pricing works
Anthropic API models are generally priced according to token usage. Input and output tokens have separate prices.
Input tokens are the information you send to Claude. They can include your prompt, system instructions, conversation history, source documents, retrieved knowledge, code, and tool-related context.
Output tokens are the content Claude generates in response.
For a document-analysis application, input cost may dominate because every request contains a large amount of source material. For code generation, long reports, or detailed reasoning tasks, generated output can represent a larger part of the cost.
The model with the lowest token price is not automatically the least expensive model for the complete workflow. If a cheaper model needs several attempts to finish a difficult task while a more capable model completes it once, the more capable model may have the lower cost per successful result.
Prompt caching can matter for repeated context
Anthropic supports prompt caching for compatible models and configurations. Caching can reduce the cost of repeatedly sending the same or similar large context, which can be important for agents, long conversations, codebase context, and applications that reuse substantial instructions or reference material.
This means two applications using the same Claude model can have very different effective costs depending on how they structure requests.
The base input and output prices in the EidoStack model table are useful for comparison, but production cost should be measured using the actual request pattern of your application.
What a context window means
The context window is the maximum amount of information Claude can work with in a single request.
That context can contain text, code, messages, documents, retrieved information, tool results, and other supported inputs. A larger context window is particularly useful when the model needs to reason over large codebases, long documents, extensive research material, or accumulated agent state.
However, context capacity is not the same as context quality. Sending more information does not automatically improve the result.
For many tasks, a smaller and more focused context is better because it reduces cost and makes the relevant information clearer.
What max output means
Max output is the maximum amount of content the model can generate in one response.
This matters for applications that need long reports, extensive code generation, large structured outputs, or other responses that can exceed ordinary chat lengths.
A large context window and a large maximum output solve different problems: context determines how much information the model can read for a request, while max output limits how much it can return.
What to compare between Claude models
When choosing between Anthropic models, look at these values together:
- Input price — the cost of prompts, instructions, conversation history, documents, code, and other context.
- Output price — the cost of the content Claude generates.
- Context window — how much information the model can consider in one request.
- Max output — how much content the model can generate in a single response.
- Latency — whether the model is responsive enough for your application.
- Quality — whether the model reliably completes the real task you need.
Use the model table above to create a shortlist. Then test those models with representative examples before choosing one for production.
- 01
Input price
What you pay for prompts, instructions, conversation history, documents, code, and other context sent to Claude.
- 02
Output price
What you pay for the tokens Claude generates in its response.
- 03
Context window
The maximum amount of information Claude can work with in a single request.
- 04
Max output
The maximum response length the model can generate for one request.
Evaluate before production
Test Anthropic models on your real prompts
Run the same prompt and context across Claude models in EidoStack, then compare response quality, token usage, context consumption, and estimated request cost.
Start FreeConnect your own Anthropic API key and evaluate Claude models under conditions that resemble your production workload.
Common Questions
What Anthropic models does EidoStack support?
EidoStack supports the Anthropic models listed in its shared model registry. The current catalog includes models from the Claude Fable, Opus, Sonnet, and Haiku families. The exact list changes as EidoStack adds, updates, or retires provider integrations.
What is the best Anthropic model?
There is no single best Claude model for every workload. Anthropic's higher-capability models are better candidates for difficult coding, reasoning, and long-running work, while Sonnet models provide a strong balance of capability, speed, and cost and Haiku models target efficient, high-volume workloads. The right choice depends on your task and quality threshold.
What is the difference between Fable, Opus, Sonnet, and Haiku?
They target different workload profiles. Fable is designed for advanced coding, knowledge work, and long-running problem solving. Opus targets complex work requiring strong reasoning and judgment. Sonnet balances capability, speed, and cost for a broad range of production workloads. Haiku prioritizes speed and cost efficiency for high-volume or well-scoped tasks.
Which Claude model is best for coding?
For difficult software-engineering and long-running coding workflows, start by evaluating current Fable, Opus, and Sonnet models available in EidoStack. The best choice depends on whether you need maximum capability or a lower-cost model that still meets your quality threshold. Test repository-level tasks rather than only short code-generation prompts.
Which Anthropic model is best for high-volume workloads?
Haiku is Anthropic's model tier designed around speed and cost efficiency. Sonnet can also be appropriate when the workload needs more capability while still operating at significant volume. Compare both on representative requests rather than choosing by price alone.
Which Anthropic models have the largest context windows?
Several recent Anthropic models in the EidoStack registry are configured with context windows of up to roughly one million tokens. The exact values shown on this page are generated from EidoStack's current AI_MODELS registry.
Does a larger context window mean a better Claude model?
No. A context window tells you how much information a model can accept in a request. It does not measure reasoning quality, reliability, latency, or how effectively the model uses that information. Large contexts can also increase request cost.
How much do Anthropic models cost?
Anthropic models have different input and output token rates, and effective cost can also be affected by features such as prompt caching and batch processing. The model table on this page shows the base values currently configured in EidoStack. Verify Anthropic's current pricing documentation before making final production estimates.
What is prompt caching?
Prompt caching allows compatible Claude requests to reuse context that has already been processed instead of paying the full standard input cost every time. It can be useful when an application repeatedly sends large system prompts, documents, code, conversation context, or other mostly unchanged information.
Can I compare Anthropic models side by side?
Yes. EidoStack is designed for model evaluation and comparison. You can run the same prompt against supported Claude models and compare their responses together with token usage, context consumption, and estimated cost.
Do I need my own Anthropic API key?
Yes. When using Anthropic models through EidoStack, you connect your own Anthropic API key. Provider-side billing, rate limits, and access to individual models are determined by your Anthropic account.
Why should I test several Claude models instead of using the most capable one?
The most capable model may be unnecessary for many workloads. A Sonnet or Haiku model may satisfy the same production quality threshold with lower latency and lower cost. Direct evaluation helps identify the least expensive model that reliably completes your real task.
How should I migrate to a newer Claude model?
Build or maintain a representative evaluation set from real application prompts. Run the old and replacement models under equivalent conditions, compare quality, latency, token usage, and cost, and examine difficult cases for regressions. Change the production model only after the replacement meets your acceptance criteria.
Official Anthropic References
Provider information is based on Anthropic's public company, model, transparency, and developer documentation. Model specifications shown in the provider table must come from EidoStack's shared AI_MODELS registry rather than duplicated Markdown data.
Last updated