Claude
Made by: Anthropic Type: Frontier AI Model Best for: Complex reasoning, long documents, nuanced writing, safe enterprise use
What it is
Claude is Anthropic's family of large language models. When you give it a task — answer a question, analyse a document, write code, summarise a meeting — it generates a response by predicting what text should come next, informed by billions of examples of human writing and reasoning.
Think of Claude less like a search engine (which retrieves facts) and more like a very knowledgeable colleague who can reason about what you give it.
The model family
| Model | Speed | Context | Best for |
|---|---|---|---|
| Claude Haiku | ⚡ Fast | 200K tokens | High-volume tasks, classification, extraction |
| Claude Sonnet | ⚖️ Balanced | 200K tokens | Most production workloads |
| Claude Opus | 🧠 Deep | 200K tokens | Complex multi-step reasoning, research |
A 200K token context window means Claude can read roughly 150,000 words in a single conversation — that's an entire book, a quarter's worth of contracts, or a full codebase.
What makes it different
Extended thinking — Claude Sonnet and Opus can "think before answering." For hard problems, they work through reasoning steps internally before producing a response. This dramatically improves accuracy on maths, logic, and multi-step tasks.
Constitutional AI — Anthropic trains Claude to be honest and avoid harmful outputs. This matters in enterprise settings where a hallucinating agent can cause real damage.
Tool use — Claude can call external tools (web search, code execution, APIs, databases) and reason about the results before continuing. This is the foundation of agentic behaviour.
Computer use (beta) — Claude Sonnet can control a computer — clicking, typing, navigating browsers — making it possible to automate tasks in any legacy system with a GUI, no API required.
200K tokens — what does that actually mean?
Token ≈ ¾ of a word. 200,000 tokens ≈ 150,000 words ≈ 500 pages.
Practical examples:
- 📄 A 50-page contract: ~25K tokens — fits easily
- 📊 A year of weekly reports: ~100K tokens — fits in one call
- 📚 War and Peace: ~580K tokens — would need chunking
Typical enterprise use cases
- Document analysis — read a 100-page RFP and pull out key obligations, deadlines, risks
- Contract review — compare against standard terms, flag deviations
- Research synthesis — take 20 analyst reports and produce an executive summary
- Customer support — read full conversation history and draft nuanced replies
- Code review — analyse a PR with full context of the surrounding codebase
How to access Claude
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: YOUR_KEY" \
-H "anthropic-version: 2023-06-01" \
-d '{
"model": "claude-sonnet-4-5",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Summarise this contract..."}]
}'Get your API key at console.anthropic.com (opens in a new tab)
Pricing at a glance
Anthropic charges per token (input + output). As a rough guide:
- Haiku — lowest cost, best for high-volume pipelines
- Sonnet — mid-tier, the most popular for production
- Opus — highest cost, use when quality is paramount
Check anthropic.com/pricing (opens in a new tab) for current rates.
Watch your output tokens. Input tokens (what you send) are cheaper than output tokens (what Claude generates). For tasks like classification or extraction, prompt Claude to give short structured answers to keep costs down.