TokenMix Research Lab · 2026-04-12

Claude API Tutorial 2026: Sonnet 4.6, Cache, Tools, Routing

Last Updated: 2026-04-30
Author: TokenMix Research Lab
Data checked: 2026-04-30

Start new Claude API projects on Sonnet 4.6. Use Haiku 4.5 for cheap routing, Opus 4.7 for hard tasks, prompt caching for repeated context, and batch for async jobs.

The Claude API is no longer a three-model "Haiku 3.5 / Sonnet 4 / Opus 4" stack. Current developer work should start from the live Claude Platform: Messages API, model deprecations, pricing, prompt caching, tool use, streaming, OpenAI SDK compatibility, and rate limits. This tutorial gives the minimum production path without old model IDs.

For broader budget planning, compare Claude against OpenAI, Gemini, DeepSeek, Grok, and Kimi in the LLM API Pricing 2026 guide.

Table of Contents

• Quick Verdict
• Current Claude API Model Table
• Step 1: Create A Console Project And API Key
• Step 2: First Python Call
• Step 3: First TypeScript Call
• Step 4: Messages API Format
• Step 5: Streaming
• Step 6: Tool Use
• Step 7: Prompt Caching
• Step 8: OpenAI SDK Compatibility
• Step 9: Use TokenMix.ai For Routing
• Cost Control Checklist
• Common Errors
• Final Recommendation
• FAQ
• Related Articles
• Sources

Quick Verdict

Use native Anthropic SDK when you need Claude-specific features. Use OpenAI SDK compatibility or TokenMix.ai when migration speed, multi-model routing, and fallback matter more.

Current Claude API Model Table

Avoid retired/deprecated defaults: Claude 3.7 Sonnet is retired, Sonnet 4 is deprecated, and Haiku 3.5 is retired. If an old tutorial tells you to start with those models, update it before shipping.

Step 1: Create A Console Project And API Key

Do not assume every new account receives a fixed free credit amount. Check the balance shown in Console. For more on credits, see our Claude API free tier guide.

Install SDKs:

pip install anthropic
npm install @anthropic-ai/sdk

Set the key:

export ANTHROPIC_API_KEY="sk-ant-your-key"

Step 2: First Python Call

from anthropic import Anthropic

client = Anthropic()

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=512,
    messages=[
        {"role": "user", "content": "Explain prompt caching in two practical bullets."}
    ],
)

print(message.content[0].text)

Use Sonnet 4.6 for the first real quality test. Use Haiku 4.5 for basic integration checks where answer quality does not matter.

Step 3: First TypeScript Call

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic();

const message = await client.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 512,
  messages: [
    { role: "user", content: "Write a concise API onboarding checklist." },
  ],
});

const first = message.content[0];
if (first.type === "text") {
  console.log(first.text);
}

Use environment variables in production. Do not commit API keys or paste them into client-side code.

Step 4: Messages API Format

Claude's native Messages API is close to chat-completions patterns, but not identical.

System prompt example:

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=700,
    system="You are a senior backend engineer. Answer with implementation details.",
    messages=[{"role": "user", "content": "How should I retry 429 errors?"}],
)

Step 5: Streaming

Use streaming when answers may be long or user-facing latency matters.

from anthropic import Anthropic

client = Anthropic()

with client.messages.stream(
    model="claude-sonnet-4-6",
    max_tokens=1200,
    messages=[{"role": "user", "content": "Explain API gateway fallback design."}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

    final = stream.get_final_message()
    print(f"\nOutput tokens: {final.usage.output_tokens}")

Step 6: Tool Use

Claude can choose tools by emitting tool_use blocks. Your application executes the tool, then returns a tool_result.

from anthropic import Anthropic

client = Anthropic()

tools = [
    {
        "name": "get_order_status",
        "description": "Look up an order by ID.",
        "input_schema": {
            "type": "object",
            "properties": {
                "order_id": {"type": "string", "description": "Order ID"}
            },
            "required": ["order_id"],
        },
    }
]

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=600,
    tools=tools,
    messages=[{"role": "user", "content": "Where is order A123?"}],
)

for block in response.content:
    if block.type == "tool_use":
        print(block.name, block.input)

Step 7: Prompt Caching

For exact cache-read, 5-minute write, and 1-hour write cost math, use the Claude API cache pricing guide.

Prompt caching is one of Claude's best cost controls. Use it when a large prefix repeats across requests: policy text, tool schemas, long documents, project context, or retrieval bundles.

POLICY = "Long repeated policy or product documentation..."

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=600,
    system=[
        {
            "type": "text",
            "text": POLICY,
            "cache_control": {"type": "ephemeral"},
        }
    ],
    messages=[{"role": "user", "content": "Answer using the policy."}],
)

usage = message.usage
print("cache write", usage.cache_creation_input_tokens)
print("cache read", usage.cache_read_input_tokens)

For deeper economics, use our Anthropic API pricing guide.

Step 8: OpenAI SDK Compatibility

Anthropic now documents an OpenAI SDK compatibility layer for testing Claude with the OpenAI SDK. This is useful for migration experiments, but the native Anthropic SDK still exposes Claude-specific behavior more directly.

If you need multi-provider routing, gateway observability, and fallback, TokenMix.ai is the more complete production pattern.

Step 9: Use TokenMix.ai For Routing

TokenMix.ai lets you call Claude and 300+ other models through an OpenAI-compatible endpoint. Use it when the product needs model choice, not one-provider purity.

from openai import OpenAI

client = OpenAI(
    api_key="your-tokenmix-key",
    base_url="https://api.tokenmix.ai/v1",
)

response = client.chat.completions.create(
    model="claude-sonnet-4-6",
    messages=[
        {"role": "system", "content": "You are a concise API tutor."},
        {"role": "user", "content": "Explain retries for 529 errors."},
    ],
)

print(response.choices[0].message.content)

Cost Control Checklist

Example monthly cost for 100M input and 30M output:

Common Errors

Use our Claude Rate Exceeded guide and Claude 529 guide for production retry design.

Final Recommendation

Build new Claude API apps on Sonnet 4.6, cache repeated context, batch async work, and route simple tasks away from Sonnet. Use TokenMix.ai when fallback and multi-model cost control matter.

FAQ

What Claude model should I use first?

Use Sonnet 4.6 for real quality tests. Use Haiku 4.5 for cheap SDK smoke tests. Use Opus 4.7 only when Sonnet fails or the task is high-value.

Does Claude API have free credits?

Sometimes an account may show onboarding or promotional credits, but there is no permanent free Claude API tier. Check your actual Claude Console balance.

Is claude-sonnet-4-20250514 still a good tutorial model?

No. It is deprecated and scheduled to retire. Use claude-sonnet-4-6 for new examples.

Can I use the OpenAI SDK with Claude?

Yes for compatibility testing through Anthropic's documented compatibility layer, and also through TokenMix.ai's OpenAI-compatible gateway. Use the native Anthropic SDK for the most Claude-specific features.

How much does prompt caching save?

Cache reads are priced at 10% of base input tokens. Actual savings depend on cache hit rate, context size, and how often the repeated prefix changes.

Should I use Batch API?

Yes for async workloads such as bulk summarization, enrichment, evaluation, and offline report generation. Batch pricing gives a 50% token discount.

What is the biggest Claude API cost mistake?

Sending every task to Sonnet or Opus. Route simple work to Haiku or cheaper non-Claude models, and reserve premium models for tasks where quality changes the outcome.

How does TokenMix.ai fit into a Claude API stack?

TokenMix.ai is the routing layer. Use it to call Claude, compare alternatives, track spend, and fail over when Claude is rate-limited or overloaded.

Related Articles

• Claude API Cache Pricing 2026: 90% Input Savings Explained
• Claude API Pricing 2026: Opus, Sonnet, Haiku Costs Compared
• Anthropic API Pricing 2026: Cache, Batch, Data Residency Fees
• Claude Rate Exceeded Error 2026: 5 Fixes for 429 Limits
• Claude API Error 529 2026: Overload Retry and Failover Guide
• Claude Sonnet vs Opus 2026: Pricing, Quality, Routing Guide
• Claude Haiku vs Sonnet 2026: Cost, Quality, Routing Rules
• OpenAI-Compatible API Gateway: 9 Providers, One SDK Guide

Sources

• Claude Messages API examples
• Claude API pricing
• Claude model deprecations
• Claude prompt caching
• Claude tool use
• Claude streaming
• Claude OpenAI SDK compatibility
• Claude API rate limits