TL;DR
Claude Mythos (internal codename “Capybara”) appeared in accidentally exposed Anthropic documents. Reported to achieve “dramatically higher scores” than Opus 4.6 on coding, academic reasoning, and cybersecurity. No public access, no published pricing, no release timeline. Build with Claude Opus 4.6 now — it’s fully available, well-documented, and any prompts and architecture you build today will transfer to Mythos when it releases.
Introduction
In early 2026, Fortune reported on Anthropic documents that were accidentally exposed, containing draft information about a model codenamed “Claude Mythos” (internally “Capybara”). The information was unverified draft content, not an official announcement.
This guide covers what was reported, what’s actually known versus speculated, and how developers should respond.
What Claude Opus 4.6 delivers today
Before evaluating Mythos, understand what the current frontier model already provides:
Coding performance:
- 65.4% on Terminal-Bench 2.0
- 72.7% on OSWorld
- 80.9% on SWE-bench Verified — the highest published score as of early 2026
API access:
- Full production API through Anthropic
- 1 million token context window at standard pricing
- 67% cost reduction from earlier versions
- Pricing: $5 input / $25 output per million tokens
Capabilities:
- Complex multi-file code generation and refactoring
- Autonomous debugging loops
- Long-document analysis and synthesis
- Computer use (controlling UIs programmatically)
What the Mythos leak said
The accidentally exposed Anthropic documents reportedly contained:
Claimed performance:“Dramatically higher scores” than Opus 4.6 on:
- Coding benchmarks
- Academic reasoning
- Cybersecurity tasks
Positioning:Described as a “new tier above Opus models” rather than an incremental version update. This language suggests it’s positioned as a different class of capability.
Cybersecurity:Noted as “currently far ahead of any other AI model in cyber capabilities.” This is the most specific capability claim in the reports.
Access:Expected to be expensive to operate. Early access limited to “cyber defense organizations” specifically.
What remains unknown
Everything significant about Mythos is unknown:
- Pricing: No published numbers. “Expensive to run” is the only guidance.
- Release timeline: No public announcement. No date.
- Public API: No indication of when general developer access happens.
- Benchmark scores: “Dramatically higher” is a claim, not a number.
- Availability: Early access focused on cyber defense; general availability is further out.
The source was an accidentally exposed draft document, not an official announcement. Details in unfinished drafts don’t reflect final decisions.
Should you wait for Mythos?
No. Build with Claude Opus 4.6.
Three reasons:
No timeline exists. You can’t build a product roadmap around “eventually.”
Architecture transfers. Prompts, system messages, API integration patterns, and workflows built for Opus 4.6 will transfer to Mythos. Anthropic maintains backward compatibility. Building now isn’t wasted work.
Opus 4.6 is already frontier. The highest SWE-bench score published, strong multimodal capabilities, and 1M token context are production-ready today.
Building today with future upgrade in mind
For applications that need to move to a more capable model when Mythos releases:
Abstract the model ID:
MODEL_CONFIG = {
"default": "claude-opus-4-6",
"high_capability": "claude-mythos" # Future upgrade
}
model = MODEL_CONFIG.get("default")
When Mythos releases, change the configuration value. No code changes required.
Design model-agnostic prompts:
Prompts that rely on specific model quirks will require updating with any model change. Write prompts that describe what you need clearly enough that any frontier model handles them.
Implement prompt caching:
At Opus 4.6’s pricing, caching system prompts reduces costs for production applications. When Mythos releases (expected to cost more), caching becomes even more important.
Testing Claude Opus 4.6 with Apidog
POST https://api.anthropic.com/v1/messages
x-api-key: {{ANTHROPIC_API_KEY}}
anthropic-version: 2023-06-01
Content-Type: application/json
{
"model": "claude-opus-4-6",
"max_tokens": 4096,
"system": "{{system_prompt}}",
"messages": [
{
"role": "user",
"content": "{{user_message}}"
}
]
}
Add assertions:
Status code is 200
Response body has field content
Response body, field stop_reason equals "end_turn"
Response time is under 60000ms
The 60-second timeout reflects that complex Opus 4.6 tasks can take 30-60 seconds. Shorter timeouts will produce false failures on legitimate requests.
Prompt caching (for repeated system prompts):
{
"model": "claude-opus-4-6",
"max_tokens": 4096,
"system": [
{
"type": "text",
"text": "{{long_system_prompt}}",
"cache_control": {"type": "ephemeral"}
}
],
"messages": [...]
}
The cache_control field enables prompt caching. Anthropic caches the marked content and charges reduced rates for cache hits. For applications with consistent system prompts, this reduces per-request cost significantly.
FAQ
Is the Mythos information reliable?
It came from accidentally exposed Anthropic documents described as drafts. Draft documents don’t reflect final decisions. Treat it as directional information about future plans, not confirmed specifications.
When will Mythos be publicly available?
No timeline exists. Early access was focused on cyber defense organizations. General developer access has no announced date.
Does the cybersecurity focus mean Mythos won’t be useful for general development?
Early access limitations don’t indicate permanent restrictions. GPT-4 had restricted access initially and became broadly available. Anthropic’s pattern is restricted preview followed by general access.
Should I pay for Claude Opus 4.6 now if Mythos might be better?
Yes. Build what you need to build today. The Opus 4.6 pricing reduction (67% cheaper than previous versions) makes it more accessible than the previous frontier tier. Waiting for future models means not building today.
Can I sign up for Mythos early access?
Anthropic hasn’t published a public early access program for Mythos. Monitor Anthropic’s announcements for access information when it becomes available.
