Claude Mythos API: Access, Model ID, and What We Know So Far

Claude Mythos is not yet available through the public Claude API, Claude.ai, or any third-party platform as of March 28, 2026. The model — Anthropic’s most powerful ever, sitting in the new Capybara tier above Opus — is currently restricted to select cyber defense organizations for early testing. Here is everything developers need to know about the Claude Mythos API: expected model IDs, pricing, migration path, and release timeline.

Claude Mythos API — model architecture and access overview

What Is Claude Mythos?

The Capybara Tier Explained

Anthropic’s model hierarchy has operated on three tiers since 2023: Haiku (fast and cheap), Sonnet (balanced), and Opus (most capable). Claude Mythos introduces Capybara as a fourth tier above Opus, representing what Anthropic calls a “step change” rather than an incremental version bump. Think of it as the difference between moving from GPT-3.5 to GPT-4 — not just better scores, but qualitatively different capabilities.

The Capybara tier is designed for tasks where maximum intelligence matters more than cost or speed. Anthropic’s leaked materials describe it as “larger and more intelligent than our Opus models,” with particular strength in cybersecurity, complex reasoning, and multi-step agent workflows.

How Mythos Differs from Opus 4.6

Claude Opus 4.6 is currently Anthropic’s publicly available flagship model, priced at $5 per million input tokens and $25 per million output tokens. Mythos achieves “dramatically higher scores” than Opus 4.6 across three key benchmarks: software coding with Mythos, academic reasoning, and cybersecurity.

The gap is not marginal. Internal documents describe Mythos as “far ahead of any other AI model in cyber capabilities” and characterize the improvement as an entirely new tier of performance. Opus 4.6 is already the best model many developers have used — Mythos reportedly makes it look like last generation’s technology in certain domains.

Current Claude API Architecture

Available Models and Model IDs

The Claude API currently offers these models through the Messages API:

ModelModel IDInputOutputContext
Claude Opus 4.6claude-opus-4-6-20250415$5/MTok$25/MTok200K (1M available)
Claude Opus 4.5claude-opus-4-5-20250220$5/MTok$25/MTok200K (1M available)
Claude Sonnet 4.6claude-sonnet-4-6-20250514$3/MTok$15/MTok200K (1M available)
Claude Sonnet 4.5claude-sonnet-4-5-20241022$3/MTok$15/MTok200K (1M beta)
Claude Haiku 4.5claude-haiku-4-5-20251001$1/MTok$5/MTok200K

Every model uses the same Messages API structure. The only parameter that changes when switching models is the model string — your prompts, tools, and response handling stay identical.

Messages API Structure

All Claude models share the same API endpoint and request format. A typical API call looks like this:

import anthropic

client = anthropic.Anthropic()
message = client.messages.create(
    model="claude-opus-4-6-20250415",
    max_tokens=4096,
    messages=[
        {"role": "user", "content": "Analyze this codebase for vulnerabilities"}
    ]
)

When Mythos becomes available, you will only need to change the model parameter to the Capybara model ID. The rest of your code — authentication, message format, tool definitions, streaming handlers — remains untouched.

Authentication and Rate Limits

The Claude API uses API key authentication via the x-api-key header or the ANTHROPIC_API_KEY environment variable. Rate limits vary by usage tier (Tier 1 through Tier 4, plus Enterprise), and Anthropic has not yet announced what tier requirements or rate limits will apply to Capybara-tier models.

Current Opus 4.6 rate limits max out at 4,000 requests per minute at Tier 4. Given that Capybara models are described as “very expensive for us to serve,” initial rate limits will likely be more restrictive than current Opus limits.

Claude Mythos API Access: Current Status

Who Has Access Now

As of March 28, 2026, Claude Mythos is available only to a small group of early-access customers selected by Anthropic. These are primarily cyber defense organizations tasked with evaluating the model’s security applications. Access is not available through the standard API console, Claude.ai, or any third-party platform.

Anthropic confirmed the model’s existence after the March 27 leak, stating it is “the most capable we’ve built to date” and is undergoing a “cautious rollout” starting with security-focused partners.

Early Access Program

Anthropic has not published a formal early access application process for Claude Mythos. Based on the leaked materials and Anthropic’s public statements, the initial rollout prioritizes organizations working in cybersecurity defense — companies and government agencies that can help validate the model’s capabilities while building safeguards against misuse.

Developers interested in early access should monitor Anthropic’s official channels and consider contacting the enterprise sales team to express interest. Previous Anthropic early access programs for features like extended context windows and tool use started with Tier 4 API customers and enterprise accounts.

Expected Public Release Timeline

Anthropic has not committed to a specific public release date for Claude Mythos. However, several factors point toward a likely timeline. Anthropic is reportedly considering an IPO in October 2026 at a $60 billion+ valuation, and launching their most impressive model before that event would maximize market impact. Safety evaluations — not commercial schedules — will determine the actual release, but a mid-to-late 2026 public API launch is the consensus estimate among industry analysts.

The release will likely follow a staged pattern: restricted early access (current phase), expanded early access to more enterprise customers, general API availability, and finally integration into Claude.ai consumer products.

Preparing for Claude Mythos API

How the Model Parameter Works

The Claude API’s design makes model upgrades straightforward. Every API request includes a model parameter that specifies which Claude model processes the request. Switching to Capybara will require changing a single string in your API calls — no architectural changes, no new endpoints, no different authentication.

This is by design. Anthropic built the Messages API to be model-agnostic, so developers can upgrade (or downgrade) models with minimal code changes. The same system prompt, tool definitions, and conversation history work across all model tiers.

Switching from Opus to Capybara

For developers currently using Claude Opus 4.6, the migration path to Claude Mythos will be simple at the code level:

# Current: Claude Opus 4.6
message = client.messages.create(
    model="claude-opus-4-6-20250415",
    max_tokens=4096,
    messages=messages
)

# Future: Claude Mythos (Capybara)
# Model ID is speculative — actual ID will be announced by Anthropic
message = client.messages.create(
    model="claude-capybara-1-0-20260801",  # placeholder
    max_tokens=4096,
    messages=messages
)

The real preparation is budgetary, not technical. Capybara-tier pricing is expected to be 2-5x higher than Opus. If your application currently spends $1,000/month on Opus API calls, budget $2,000-$5,000/month for equivalent Capybara usage. Run cost projections now so the pricing doesn’t force a last-minute architecture change.

Code Examples for Future Migration

Smart developers are already building their applications with model flexibility in mind. Use environment variables or configuration files for model selection:

import os
import anthropic

MODEL = os.getenv("CLAUDE_MODEL", "claude-opus-4-6-20250415")
client = anthropic.Anthropic()

def analyze(prompt: str) -> str:
    message = client.messages.create(
        model=MODEL,
        max_tokens=8192,
        messages=[{"role": "user", "content": prompt}]
    )
    return message.content[0].text

This pattern lets you switch to Mythos on launch day by updating a single environment variable — no code deployment needed. For agent workflows using tool use, the same approach applies since Capybara will support the same tool use protocol.

Expected Claude Mythos API Specifications

Predicted Model ID Format

Anthropic follows a consistent naming convention: claude-{tier}-{version}-{date}. Current examples include claude-opus-4-6-20250415 and claude-sonnet-4-6-20250514. The Mythos model ID will likely follow the pattern claude-capybara-{version}-{date}, though Anthropic could choose to use the “mythos” name directly, such as claude-mythos-1-0-{date}.

There is precedent for both approaches. The tier name (capybara) would be consistent with haiku/sonnet/opus naming, while the model name (mythos) would mirror how specific model identifiers work. Anthropic has not confirmed either format.

Context Window Expectations

Claude Opus 4.6 and Sonnet 4.6 both offer 1 million token context windows at standard pricing — a feature introduced with the 4.6 generation. It would be unusual for Anthropic to reduce this for their most capable model, so a 1M token context window is the minimum expectation for Mythos.

Given that Capybara is positioned as a breakthrough-tier model, there is speculation that it could offer an even larger context window — potentially 2M or more tokens. However, no leaked materials have confirmed context window specifications.

Pricing Estimates

Anthropic has acknowledged that Mythos is “very expensive for us to serve, and will be very expensive for our customers to use.” Community estimates for Capybara-tier API pricing fall into two ranges:

EstimateInputOutputMultiplier vs Opus
Conservative$10-15/MTok$50-75/MTok2-3x
Aggressive$20-25/MTok$100-125/MTok4-5x

The batch API discount of 50% will likely apply to Capybara, making batch processing the cost-effective choice for non-time-sensitive workloads. At conservative estimates, batch pricing would be $5-7.50 input and $25-37.50 output per million tokens — comparable to current Opus standard pricing.

Prompt caching will also be critical for managing Capybara costs. Current Opus caching reduces cache hits to 10% of the standard input price, and the same multiplier on Capybara pricing would make cached requests significantly more affordable.

Claude Mythos on Third-Party Platforms

AWS Bedrock Availability

Claude models are available on AWS Bedrock, and every major Claude release has eventually appeared there. AWS Bedrock currently offers Claude Opus 4.6, Sonnet 4.6, and Haiku 4.5 with both global and regional endpoints. Capybara will almost certainly follow, but expect a delay between the Anthropic API launch and Bedrock availability — typically 2-6 weeks for new model tiers.

Regional endpoints on Bedrock include a 10% pricing premium over global endpoints, a structure that applies to all 4.5+ models. Capybara pricing on Bedrock will include this premium on top of already-higher base pricing.

Google Vertex AI

Google Vertex AI mirrors the Bedrock model, offering Claude models with both global and regional endpoints. The same 10% regional premium applies. Vertex AI integration typically follows a similar timeline to Bedrock, with new models appearing weeks after the Anthropic API release.

Microsoft Foundry

Microsoft Foundry (formerly Azure AI Foundry) also hosts Claude models. For enterprise customers already using Microsoft’s cloud infrastructure, this provides an alternative deployment path. Pricing on Foundry is set separately from Anthropic’s direct pricing — check Microsoft’s pricing page for details when Capybara becomes available.

Questions About Claude Mythos API

How do I access the Claude Mythos API right now?

You cannot access Claude Mythos through any public API as of March 28, 2026. The model is restricted to select cyber defense organizations in an early access program. Monitor Anthropic’s official channels and contact their enterprise sales team to express interest in early access.

What will the Claude Mythos model ID be?

Anthropic has not announced the model ID. Based on their naming convention, it will likely follow the format claude-capybara-{version}-{date} or claude-mythos-{version}-{date}. The exact ID will be published when API access opens.

When will Claude Mythos be publicly available?

No confirmed date. Safety evaluations determine the timeline. Industry analysts expect a mid-to-late 2026 public API launch, potentially timed ahead of Anthropic’s reported October 2026 IPO plans.

How much will the Claude Mythos API cost?

Anthropic confirmed it will be “very expensive.” Conservative estimates: $10-15 per million input tokens and $50-75 per million output tokens (2-3x current Opus pricing). Aggressive estimates go up to 4-5x Opus.

Do I need to change my code to use Claude Mythos?

No. The Claude Messages API is model-agnostic. You only need to change the model parameter string in your API calls. All other code — authentication, prompts, tool definitions, streaming — stays the same.

Will Claude Mythos support tool use and agents?

Yes. The Capybara tier will support the same tool use protocol as current Claude models, including computer use, bash execution, and custom tool definitions. Mythos is specifically designed for “improved consistency in agent workflows.”

Will Claude Mythos be available on AWS Bedrock and Google Vertex AI?

Almost certainly, but expect a delay of 2-6 weeks after the Anthropic API launch. All major Claude models have eventually appeared on both platforms with both global and regional endpoint options.

What is the difference between Capybara and Opus?

Capybara is a new model tier above Opus in Anthropic’s hierarchy: Haiku (fast) → Sonnet (balanced) → Opus (flagship) → Capybara (breakthrough). Claude Mythos is the first Capybara-tier model, achieving “dramatically higher scores” than Opus 4.6 on coding, reasoning, and cybersecurity benchmarks.

keyboard_arrow_up