Anthropic Claude¶

What it is¶

Anthropic is an AI safety and research company that produces the Claude family of LLMs. As of June 2026, it is a proprietary service offering high-performance models known for strong reasoning, coding excellence, and safety. Pricing is usage-based with a free testing tier available via the Anthropic Console.

What problem it solves¶

It offers a high-performance alternative to OpenAI with a focus on "Constitutional AI" (safety) and exceptional performance in coding, long-form document analysis, and complex reasoning tasks. It provides a reliable engine for autonomous agents via native Model Context Protocol (MCP) support.

Where it fits in the stack¶

LLM / Reasoning Engine / Provider. It serves as the primary intelligence layer for coding agents and complex document synthesis workflows.

Typical use cases¶

Pair Programming: Claude 3.5 Sonnet and 4.8 Opus are the preferred models for tools like Aider.
Complex Analysis: Summarizing long technical documentation or legal files using the 200k+ token context window.
Strict Adherence: Workflows requiring high precision in following complex formatting or reasoning rules.
Autonomous Engineering: Leveraging MCP 3.0 to enable Claude to interact with local and remote tools.
Computer Use: Utilizing Claude 4.8 Opus for direct interaction with operating systems and browsers.

Model routing (June 2026)¶

Model	Primary Use Case	Default?
Haiku	Fast classification, extraction, rewriting, and cheap high-volume tasks	No
Sonnet	Default coding, planning, and most daily serious work	Yes
Opus 4.7	Complex software engineering and high-resolution vision tasks	No
Opus 4.8	Premium escalation for hard synthesis and autonomous browser-agent tasks	No
Mythos	Frontier-scale simulations and high-reliability software factory architectures	No

Strengths¶

Coding Excellence: Widely regarded as one of the strongest daily-driver models for software engineering.
Safety Focus: Built with Constitutional AI principles for better alignment and reduced harmful outputs.
Large Context: Ability to handle up to 2.5M tokens in Plandex integrations.
Low Hallucination: Exhibits high factual accuracy and honesty in complex reasoning.
Native MCP Support: Seamless integration with the Model Context Protocol (MCP 3.0) for extensible tool use.

Limitations¶

Cloud Dependency: Requires external API access; no official local/offline version.
Rate Limits: Usage tiers can be restrictive for new accounts.
Cost: High-end models like Opus 4.8 are significantly more expensive than smaller models.

When to use it¶

For software development tasks where Sonnet/Opus is the right default.
When safety and alignment are critical priorities for your application.
For analyzing very long documents or entire codebases in a single context.
When you want a multi-tier routing strategy using the Model Routing Guide.

When not to use it¶

When a local/offline solution is required for privacy or cost (consider Llama 4 Maverick).
If you need native DALL-E 3 style image generation in the same API call.

Getting started¶

Installation¶

Install the official Python SDK:

pip install anthropic

Initial Configuration¶

Set your API key as an environment variable:

export ANTHROPIC_API_KEY='your-api-key-here'

CLI examples¶

Using Claude Code¶

Claude plugins and CLI tools often interact directly with the API:

claude "Analyze the current directory and suggest refactorings"

Listing Models via SDK¶

While not a direct CLI, simple scripts can be used to check availability:

import anthropic
print(anthropic.Anthropic().models.list())

API examples¶

Basic Message Creation (Python)¶

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-4-8-opus-20260528",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain the advantages of MCP 3.0."}
    ]
)
print(message.content)

Streaming Responses¶

with client.messages.stream(
    model="claude-3-5-sonnet-latest",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Write a 500-word essay on AI safety."}]
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

OpenAI — Primary competitor for frontier models.
OpenRouter — Unified API access to Claude and others.
Aider — Popular CLI tool optimized for Claude.
MCP — Standard for extending Claude's capabilities.
Claude Code — Anthropic's agentic coding CLI.
Model Routing Guide — Strategy for model selection.
Agentic Workflows — Patterns for autonomous execution.
Plandex — Complex engineering tool supporting large context Claude models.
Zed — Editor with native Claude integration.

Sources / References¶

Contribution Metadata¶

Last reviewed: 2026-06-28
Confidence: high