Claude API

Building LLM-Powered Applications with Claude

Defaults

• Model: Claude Opus 4.6 (claude-opus-4-6) unless user requests otherwise
• Thinking: thinking: {type: "adaptive"} for anything remotely complicated
• Streaming: default for long input/output or high max_tokens

Current Models

ALWAYS use claude-opus-4-6 unless user explicitly names a different model. Use exact model ID strings — never append date suffixes.

Which Surface to Use?

Language Detection

Check project files:

• *.py, requirements.txt → Python
• *.ts, *.tsx, package.json → TypeScript
• *.js, *.jsx → TypeScript (same SDK)
• *.java, pom.xml → Java
• *.go, go.mod → Go
• *.rb, Gemfile → Ruby

Thinking & Effort

Opus 4.6 / Sonnet 4.6: Use thinking: {type: "adaptive"}. budget_tokens is deprecated.

Effort parameter: output_config: {effort: "low"|"medium"|"high"|"max"} (inside output_config). Default is high. max is Opus 4.6 only.

Older models (only if explicitly requested): thinking: {type: "enabled", budget_tokens: N}.

Quick Start (Python)

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-4-6",
    max_tokens=16000,
    thinking={"type": "adaptive"},
    messages=[{"role": "user", "content": "Your prompt here"}]
)

Quick Start (TypeScript)

import Anthropic from '@anthropic-ai/sdk'

const client = new Anthropic()

const message = await client.messages.create({
  model: 'claude-opus-4-6',
  max_tokens: 16000,
  thinking: { type: 'adaptive' },
  messages: [{ role: 'user', content: 'Your prompt here' }]
})

Tool Use Pattern

tools = [{
    "name": "get_weather",
    "description": "Get current weather for a location",
    "input_schema": {
        "type": "object",
        "properties": {
            "location": {"type": "string", "description": "City name"}
        },
        "required": ["location"]
    }
}]

message = client.messages.create(
    model="claude-opus-4-6",
    max_tokens=4096,
    tools=tools,
    messages=[{"role": "user", "content": "What's the weather in Jakarta?"}]
)

Common Pitfalls

• Don't use budget_tokens on Opus 4.6 / Sonnet 4.6 — use adaptive thinking
• Don't lowball max_tokens — non-streaming: ~16000, streaming: ~64000
• Opus 4.6 supports up to 128K max_tokens (requires streaming)
• Use output_config: {format: {...}} not deprecated output_format
• Always parse tool inputs with json.loads() / JSON.parse() — never raw string matching
• Use SDK types (Anthropic.MessageParam, Anthropic.Tool) — don't redefine
• Don't truncate inputs silently — notify user and discuss options