FAQ
Contact Us
About

Claude Capybara API Access: Current Status and How to Prepare

Claude Capybara is not yet available through public API access. Anthropic confirmed that the model is being tested with “a small group of early access customers” focused on cybersecurity defense. For everyone else, access will come later — but the Claude API uses a unified interface that makes switching to Capybara a single parameter change once it becomes available. We cover this further in our agent workflow capabilities article.

Claude Capybara API access — developer setup guide

This guide covers the current access status, how to prepare your API integration, and what developers need to know about the timeline for broader availability.

Current Access Status

As of March 2026, Claude Capybara access is restricted to a handpicked group of organizations. Understanding who has access and why explains when everyone else will get it.

Who Has Access Now

Anthropic is providing early access exclusively to cybersecurity defense organizations. This is not a typical beta program where developers sign up and wait. The selection is strategic — Anthropic wants defensive security teams to understand the model’s capabilities and develop countermeasures before Capybara becomes broadly available.

This approach is unprecedented in the AI industry. No previous frontier model has been restricted specifically because of cybersecurity concerns. It signals that Anthropic considers the gap between Capybara and existing models significant enough to require a defensive head start.

Why Not Open Access Yet

Anthropic’s internal assessment states the model “poses unprecedented cybersecurity risks” and is “currently far ahead of any other AI model in cyber capabilities.” Releasing it broadly without giving defenders time to prepare would be irresponsible by Anthropic’s own safety standards.

The company is being “deliberate” about the release — choosing safety over competitive pressure, even though the model’s existence is now public knowledge after the March 26 data leak.

How the Claude API Works

When Capybara becomes available, accessing it will require no fundamental changes to existing Claude API integrations. Anthropic’s unified API design makes model switching trivial.

Unified API Interface

Every Claude model — from Haiku to Opus — uses the same API endpoint, the same authentication, the same request format, and the same SDK. The only difference between calling Opus and calling Capybara will be the model parameter.

Current Opus 4.6 request:

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-4-6-20260220",
    max_tokens=4096,
    messages=[
        {"role": "user", "content": "Analyze this codebase for security vulnerabilities"}
    ]
)

When Capybara launches, the only change will be the model string:

message = client.messages.create(
    model="claude-capybara-...",  # model ID TBD
    max_tokens=4096,
    messages=[
        {"role": "user", "content": "Analyze this codebase for security vulnerabilities"}
    ]
)

Everything else — authentication, message format, response structure, tool use, streaming — remains identical.

Setting Up API Access Now

If you don’t already have Claude API access, setting it up now means you’ll be ready when Capybara launches. The process takes minutes.

Step 1: Create an account at console.anthropic.com.

Step 2: Generate an API key in the Console under Settings → API Keys.

Step 3: Install the Python SDK:

pip install anthropic

Step 4: Set your API key as an environment variable:

export ANTHROPIC_API_KEY="your-key-here"

Step 5: Test with a current model to confirm everything works:

import anthropic

client = anthropic.Anthropic()
message = client.messages.create(
    model="claude-sonnet-4-6-20260220",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello, Claude"}]
)
print(message.content[0].text)

When Capybara becomes available, you change one line.

Third-Party Platform Access

Claude models are also available through AWS Bedrock, Google Vertex AI, and Microsoft Foundry. Capybara will likely follow the same distribution path once public access opens, though the timeline may differ from direct API availability.

If your organization already uses Claude through one of these platforms, Capybara access will appear as a new model option in the same interface you already use.

API Features That Matter for Capybara

Several API features become particularly important when working with a premium-priced model like Capybara.

Prompt Caching

With Capybara’s expected higher per-token cost, prompt caching becomes essential for cost management — see pricing guide. Caching stores previously processed prompt content and reuses it on subsequent requests at 10% of the standard input price.

For development workflows where you send the same codebase context repeatedly, caching can reduce input costs by up to 90% on cached content. This is the single most impactful cost optimization for Capybara usage.

message = client.messages.create(
    model="claude-capybara-...",
    max_tokens=4096,
    system=[{
        "type": "text",
        "text": "You are a security auditor analyzing the following codebase...",
        "cache_control": {"type": "ephemeral"}
    }],
    messages=[{"role": "user", "content": "Find all SQL injection vulnerabilities"}]
)

Batch API

The Batch API processes requests asynchronously at a 50% discount. For non-real-time tasks — security scanning across a codebase, batch code review, large-scale refactoring analysis — this halves the cost of using Capybara.

batch = client.batches.create(
    requests=[
        {
            "custom_id": "scan-file-1",
            "params": {
                "model": "claude-capybara-...",
                "max_tokens": 4096,
                "messages": [{"role": "user", "content": "Scan this file for vulnerabilities: ..."}]
            }
        }
        # ... more requests
    ]
)

Tool Use and Agent Workflows

Capybara’s enhanced agent workflow capabilities are accessed through the same tool use API that works with all Claude models. Tools defined for Opus work identically with Capybara — the model is just better at deciding when and how to use them.

message = client.messages.create(
    model="claude-capybara-...",
    max_tokens=4096,
    tools=[{
        "name": "run_security_scan",
        "description": "Run a security scan on the specified file",
        "input_schema": {
            "type": "object",
            "properties": {
                "file_path": {"type": "string", "description": "Path to file to scan"}
            },
            "required": ["file_path"]
        }
    }],
    messages=[{"role": "user", "content": "Audit this repository for security issues"}]
)

Expected Timeline for Public Access

Anthropic has not published a release timeline. Based on available information, several scenarios are plausible.

Most Likely: Q3-Q4 2026

Bloomberg reported on March 26, 2026, that Anthropic is considering an October 2026 IPO. Launching Capybara before the IPO would demonstrate both technological leadership and revenue potential. This makes a public release in the July-September 2026 window plausible — giving Anthropic time to build revenue momentum before going public.

Phased Rollout

The most likely path to general availability follows three phases. First, the current restricted early access for cybersecurity organizations. Second, expanded access for enterprise API customers, possibly with higher rate limits and custom pricing. Third, general availability through the public API, Console, and third-party platforms.

Each phase may last weeks to months, depending on how Anthropic evaluates the risk-benefit balance.

What Could Delay It

If security testing reveals risks that Anthropic cannot mitigate, the timeline extends. If a competing lab releases a model with similar cyber capabilities without restrictions, Anthropic may feel pressure to accelerate. The timeline is genuinely uncertain — Anthropic is balancing safety commitments against competitive and business pressures.

How to Prepare for Capybara Access

Developers who want to use Capybara on day one can take concrete steps now.

Build with Opus, Upgrade to Capybara

Any application built with the Claude API and Opus 4.6 will work with Capybara by changing the model parameter. Build your workflows, tools, and prompts against Opus now. When Capybara becomes available, test the switch — most applications will work without modification.

Implement Cost Controls Early

Set up usage monitoring, tiered routing, and prompt caching before Capybara launches. If you add Capybara without cost controls in place, a single misconfigured loop could generate significant charges at premium pricing.

Join the Waitlist

While Anthropic has not announced a public waitlist, enterprise customers can contact the sales team to discuss early access. Organizations working in cybersecurity defense have the strongest case for priority access.

Questions About Claude Capybara API Access

Is Claude Capybara available through the API?

Not yet. As of March 2026, Capybara access is limited to a small group of cybersecurity defense organizations selected by Anthropic. Public API access has not been announced.

How do I switch from Opus to Capybara?

Change the model parameter in your API request. The Claude API uses a unified interface — authentication, message format, tool use, and all other parameters remain identical across models.

Will Capybara be available on AWS Bedrock?

Likely yes. All current Claude models are available on AWS Bedrock, Google Vertex AI, and Microsoft Foundry. Capybara will probably follow the same distribution path, though the timeline may differ from direct API availability.

Do I need to change my code for Capybara?

No. Any application built with the Claude API works with Capybara by changing one parameter. Tools, prompts, streaming, and all other features use the same interface across all Claude models.

When will Capybara API access open to everyone?

No official date. Community estimates point to Q3-Q4 2026, potentially aligned with Anthropic’s reported IPO timeline. Enterprise customers should contact Anthropic sales for information about expanded access.

keyboard_arrow_up