Claude API Model ID List: Opus, Sonnet, and Haiku IDs for 2026

When you call Claude API, the model field must be an exact model ID string. Display names such as Claude Sonnet, shorthand names such as haiku, or malformed IDs such as claude-sonnet4.6 will fail.

This guide lists the Claude model IDs currently referenced by ClaudeAPI, explains how the naming pattern works, and shows how to route tasks across Opus, Sonnet, and Haiku in real applications.

Model availability can change. For production, always confirm the final list in the ClaudeAPI console or through the /v1/models endpoint before deploying.

Connection parameters

These values apply across most ClaudeAPI integrations:

For OpenAI-compatible SDKs and tools, pass the model ID directly in the model field:

{
  "model": "claude-sonnet-4-6"
}

{
  "model": "claude-sonnet-4-6"
}

Copy-ready model ID list

Use this section when you just need the string.

# Opus 4 family
claude-opus-4-8
claude-opus-4-7
claude-opus-4-6
claude-opus-4-5-20251101

# Sonnet 4 family
claude-sonnet-4-6
claude-sonnet-4-5-20250929

# Haiku 4 family
claude-haiku-4-5-20251001

# Opus 4 family
claude-opus-4-8
claude-opus-4-7
claude-opus-4-6
claude-opus-4-5-20251101

# Sonnet 4 family
claude-sonnet-4-6
claude-sonnet-4-5-20250929

# Haiku 4 family
claude-haiku-4-5-20251001

The Haiku ID is the one people mistype most often. The date suffix is part of the model ID:

Correct: claude-haiku-4-5-20251001
Wrong:   claude-haiku-4-5

Correct: claude-haiku-4-5-20251001
Wrong:   claude-haiku-4-5

Opus 4 models

Opus is the flagship family for the hardest work: complex refactoring, multi-step reasoning, long-document analysis, agent coordination, and high-quality generation.

Use claude-opus-4-8 for new projects when you need the strongest model in this list. Use older stable IDs only when you have already tuned prompts against them and want to reduce behavior changes.

Sonnet 4 models

Sonnet is the default choice for most production applications. It is strong enough for coding, content generation, RAG answers, data analysis, workflow automation, and general chat while staying more cost-efficient than Opus.

If you are not sure where to start, use:

claude-sonnet-4-6

claude-sonnet-4-6

Then move specific high-volume lightweight tasks to Haiku, or move only the hardest tasks to Opus.

Haiku 4 models

Haiku is best for high-frequency, lightweight, and latency-sensitive work.

Good Haiku use cases:

• routing decisions
• classification
• formatting
• short extraction
• lightweight Q&A
• background autocomplete in coding tools

The full ID matters:

claude-haiku-4-5-20251001

claude-haiku-4-5-20251001

Do not shorten it to:

claude-haiku-4-5

claude-haiku-4-5

Model selection table

A simple rule:

Not sure? Use Sonnet.
Need the highest quality? Use Opus.
Need speed and low cost at scale? Use Haiku.

Not sure? Use Sonnet.
Need the highest quality? Use Opus.
Need speed and low cost at scale? Use Haiku.

Where to enter the model ID

For Base URL configuration across these tools, see the Claude API Base URL configuration guide.

Model ID naming pattern

Claude model IDs follow a predictable structure:

claude - {family} - {version} [- {YYYYMMDD}]
  |        |          |            |
  |        |          |            optional date suffix
  |        |          version, written with hyphens
  |        opus / sonnet / haiku
  fixed prefix

claude - {family} - {version} [- {YYYYMMDD}]
  |        |          |            |
  |        |          |            optional date suffix
  |        |          version, written with hyphens
  |        opus / sonnet / haiku
  fixed prefix

Examples:

claude-opus-4-8
claude-sonnet-4-6
claude-haiku-4-5-20251001

claude-opus-4-8
claude-sonnet-4-6
claude-haiku-4-5-20251001

Three rules prevent most mistakes:

1. Use hyphens, not dots: claude-sonnet-4-6, not claude-sonnet-4.6.
2. If a date suffix exists, include the whole suffix.
3. Use the API ID, not the display name.

Common model ID mistakes

When debugging, compare the string character by character. Model IDs are not fuzzy matched.

Check available models with /v1/models

The safest way to verify what your account can use is to query the model list:

curl https://gw.claudeapi.com/v1/models \
  -H "Authorization: Bearer YOUR_CLAUDEAPI_KEY"

curl https://gw.claudeapi.com/v1/models \
  -H "Authorization: Bearer YOUR_CLAUDEAPI_KEY"

You should receive an OpenAI-style model list:

{
  "object": "list",
  "data": [
    {
      "id": "claude-haiku-4-5-20251001",
      "object": "model"
    },
    {
      "id": "claude-sonnet-4-6",
      "object": "model"
    }
  ]
}

{
  "object": "list",
  "data": [
    {
      "id": "claude-haiku-4-5-20251001",
      "object": "model"
    },
    {
      "id": "claude-sonnet-4-6",
      "object": "model"
    }
  ]
}

The id field is the exact value to paste into your API request.

Python example: route by task type

Model routing lets you control cost without making every request weaker.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["CLAUDE_API_KEY"],
    base_url="https://gw.claudeapi.com/v1",
)

def get_model(task_type: str) -> str:
    routing = {
        "outline": "claude-opus-4-8",
        "paragraph": "claude-sonnet-4-6",
        "format": "claude-haiku-4-5-20251001",
    }
    return routing.get(task_type, "claude-sonnet-4-6")

def generate(task_type: str, prompt: str) -> str:
    model = get_model(task_type)
    response = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": prompt}],
        max_tokens=2048,
    )
    return response.choices[0].message.content

outline = generate("outline", "Create an outline for an article about AI coding tools.")
paragraph = generate("paragraph", "Expand section two in about 300 words.")
formatted = generate("format", "Convert this text into Markdown: ...")

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["CLAUDE_API_KEY"],
    base_url="https://gw.claudeapi.com/v1",
)

def get_model(task_type: str) -> str:
    routing = {
        "outline": "claude-opus-4-8",
        "paragraph": "claude-sonnet-4-6",
        "format": "claude-haiku-4-5-20251001",
    }
    return routing.get(task_type, "claude-sonnet-4-6")

def generate(task_type: str, prompt: str) -> str:
    model = get_model(task_type)
    response = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": prompt}],
        max_tokens=2048,
    )
    return response.choices[0].message.content

outline = generate("outline", "Create an outline for an article about AI coding tools.")
paragraph = generate("paragraph", "Expand section two in about 300 words.")
formatted = generate("format", "Convert this text into Markdown: ...")

In this example:

• Opus handles planning and difficult reasoning.
• Sonnet handles the main generation work.
• Haiku handles formatting.

Python example: route by prompt complexity

You can also route dynamically based on input length and keywords.

def dynamic_route(user_message: str) -> str:
    estimated_tokens = len(user_message) / 4
    complex_keywords = {
        "analyze",
        "reason",
        "compare",
        "review",
        "refactor",
        "summarize full document",
        "legal",
        "compliance",
    }
    has_complex_signal = any(keyword in user_message.lower() for keyword in complex_keywords)

    if estimated_tokens > 1000 and has_complex_signal:
        return "claude-opus-4-8"
    if estimated_tokens > 200 or has_complex_signal:
        return "claude-sonnet-4-6"
    return "claude-haiku-4-5-20251001"

def smart_chat(user_message: str) -> str:
    model = dynamic_route(user_message)
    print(f"Using model: {model}")

    response = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": user_message}],
        max_tokens=2048,
    )
    return response.choices[0].message.content

def dynamic_route(user_message: str) -> str:
    estimated_tokens = len(user_message) / 4
    complex_keywords = {
        "analyze",
        "reason",
        "compare",
        "review",
        "refactor",
        "summarize full document",
        "legal",
        "compliance",
    }
    has_complex_signal = any(keyword in user_message.lower() for keyword in complex_keywords)

    if estimated_tokens > 1000 and has_complex_signal:
        return "claude-opus-4-8"
    if estimated_tokens > 200 or has_complex_signal:
        return "claude-sonnet-4-6"
    return "claude-haiku-4-5-20251001"

def smart_chat(user_message: str) -> str:
    model = dynamic_route(user_message)
    print(f"Using model: {model}")

    response = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": user_message}],
        max_tokens=2048,
    )
    return response.choices[0].message.content

In production, log the selected model, estimated input size, token usage, latency, and user satisfaction signal. Run the router for one or two weeks, then adjust thresholds with real data.

Cost estimation formula

Claude API cost is typically calculated from input and output tokens:

cost =
  input_tokens / 1,000,000 * input_price
+ output_tokens / 1,000,000 * output_price

cost =
  input_tokens / 1,000,000 * input_price
+ output_tokens / 1,000,000 * output_price

Example with the source article’s ClaudeAPI Sonnet 4.6 price:

input:  2,000 tokens * $2.4 / 1,000,000 = $0.0048
output: 1,000 tokens * $12  / 1,000,000 = $0.0120
total:                                      $0.0168

input:  2,000 tokens * $2.4 / 1,000,000 = $0.0048
output: 1,000 tokens * $12  / 1,000,000 = $0.0120
total:                                      $0.0168

Always use the console billing page for final accounting.

FAQ

Which model should I use first?

Start with claude-sonnet-4-6. It is the best default for most coding, RAG, writing, and chat applications.

Which model is best for RAG?

Use claude-sonnet-4-6 for most RAG systems. Move to Opus only when the task requires stronger cross-document reasoning or difficult synthesis.

Which model is best for autocomplete or high-volume routing?

Use claude-haiku-4-5-20251001. It is the most cost-efficient model in this list and is a good fit for lightweight, frequent calls.

Why does claude-haiku-4-5 fail?

Because the full ID includes the date suffix:

claude-haiku-4-5-20251001

claude-haiku-4-5-20251001

The shortened value is not the same model ID.

Are ClaudeAPI prices the same as Anthropic official prices?

The source article states that ClaudeAPI offers prices based on Anthropic’s public rates with a platform discount. Treat ClaudeAPI prices as platform-specific and verify them in the ClaudeAPI console before budgeting.

How do I know the current model list?

Use:

curl https://gw.claudeapi.com/v1/models \
  -H "Authorization: Bearer YOUR_CLAUDEAPI_KEY"

curl https://gw.claudeapi.com/v1/models \
  -H "Authorization: Bearer YOUR_CLAUDEAPI_KEY"

The returned data[].id values are the model IDs available to your account.

Related guides

• Claude API Base URL configuration guide
• Claude API pricing and model selection guide
• Claude API cost estimation and budget guide
• Migrate from OpenAI API to Claude API

Sources

• Anthropic models overview
• Anthropic pricing documentation
• ClaudeAPI console