Anthropic Claude

Anthropic provides Claude models with exceptional coding, reasoning, and long-context capabilities. Arcana supports prompt caching for faster, cheaper repeated contexts.

Setup

# 1. Get your API key from https://console.anthropic.com/
export ANTHROPIC_API_KEY=sk-ant-...

# 2. Verify detection
arcana doctor

# 3. Run
arcana run "hello"

Available Models

Model	Context	Best For
`claude-sonnet-4`	200K	Best balance of speed, quality, and cost
`claude-opus-4`	200K	Highest quality, complex reasoning
`claude-haiku-3.5`	200K	Fast, cheap, good for extraction

Config File

{
  "provider": "anthropic",
  "model": "claude-sonnet-4",
  "utilityModel": "claude-haiku-3.5"
}

Prompt Caching

Anthropic supports prompt caching which reduces costs for repeated context (system prompts, long documents). Arcana uses this automatically when available.

Cached input tokens cost 10% of normal price
Cache writes cost 25% more than normal input
Cache TTL is 5 minutes (extended on each use)

Environment Variables

Variable	Description
`ANTHROPIC_API_KEY`	Your Anthropic API key (required)
`ANTHROPIC_BASE_URL`	Custom base URL for proxies

Pricing Notes

Model	Input	Output
claude-sonnet-4	$3.00	$15.00
claude-opus-4	$15.00	$75.00
claude-haiku-3.5	$0.80	$4.00

Prices per 1M tokens. Check anthropic.com/pricing for current rates.

Tips

claude-sonnet-4 is the recommended default — excellent quality at reasonable cost
Use claude-haiku-3.5 as the utilityModel for extraction tasks
Claude excels at long-context tasks — 200K tokens is roughly 150K words
Prompt caching is automatic — no configuration needed

Last updated: Jul 24, 2026