On Tuesday, Anthropic released Claude Mythos 5.1, its latest restricted AI model designed for approved professionals in cybersecurity and life science research. The model can handle specialist work that would normally trigger safeguards on Anthropic's public models. It is part of a broader industry effort to serve high-stakes professional domains with frontier models while keeping dual-use capabilities — technical work that has both defensive and offensive applications — behind vetting programmes rather than open release.

Claude Mythos 5.1 costs $10 per million input tokens and $50 per million output tokens. It features a one-million-token context window and can return up to 128,000 tokens in a single response.

The model cannot be selected through a Free, Pro, Max, or Team subscription. Instead, Anthropic grants access through trusted programmes for vetted organisations, and access is currently limited to selected organisations in the United States.

What is Claude Mythos 5.1?

Claude Mythos 5.1 is Anthropic's restricted AI model for vetted cybersecurity and life science organisations. It replaces Claude Mythos 5 and offers the same general coding, research, and knowledge work capabilities as Claude Fable 5.1.

Mythos 5.1 and Fable 5.1 share the same underlying model, but the two versions differ in access rules and safeguards. Mythos applies more permissive controls for approved professional work in cybersecurity and biology, while Anthropic's other safety rules remain in place.

The model accepts text and images and returns text. It can analyse source code, research files, screenshots, tables, and document pages. Adaptive thinking is always enabled, and an effort setting controls how much reasoning the model uses.

Release date and history

Anthropic introduced Claude Mythos Preview through Project Glasswing on April 7, 2026. Selected technology organisations used it to find weaknesses in important software systems and recommend defensive fixes. Anthropic committed $100 million in usage credits to the early project and listed a post-credit price of $25 per million input tokens and $125 per million output tokens for Mythos Preview.

On June 2, Anthropic expanded Project Glasswing to about 150 more organisations across more than 15 countries. It then released Claude Mythos 5 on June 9 at a lower price of $10 for input and $50 for output per million tokens.

Export control questions disrupted Mythos 5 access shortly after launch. Anthropic said on July 1 that US government approval allowed it to restore access for a group of US organisations. The episode highlights how restricted-access AI models sit at the intersection of commercial cloud services and US export control rules, which govern where certain AI capabilities can be delivered.

Claude Mythos 5.1 launched on September 1, 2026, replacing Mythos 5 as Anthropic's newest specialist model for approved cyber defenders and life scientists. Older articles may still cite the Mythos Preview price of $25 for input and $125 for output; the current Mythos 5.1 prices are $10 and $50 per million tokens.

Pricing

Anthropic charges based on the amount of text and reasoning the model processes. Final cost depends on the prompt, the response, and how much effort and caching are used.

Prompt caching lets an organisation store information the model will reuse, such as a large code repository, research documents, or tool instructions. Each cache read costs $0.25 per million tokens. Mythos 5 charged $1 per million cache read tokens, so Mythos 5.1 cuts that figure by 75%. Cache writes remain at $12.50 for five minutes and $20 for one hour.

Batch processing costs half the normal input and output rates and suits work that does not require an immediate response.

Claude subscriptions do not provide standard Mythos 5.1 access. Free, Pro, Max, and Team users cannot choose it from the standard model selector. An organisation must receive approval through an Anthropic trusted access programme, along with separate API or cloud platform billing. Approval does not remove usage charges.

Features

Mythos 5.1 supports approved defensive cybersecurity research. It can help vetted teams identify weaknesses in source code, examine software in controlled environments, and suggest fixes for human review. Anthropic limits access because the same technical skills could be misused; the trusted programmes require organisations to use the model only for authorised, defensive work.

The model also supports advanced biology and life science research for approved professionals. Anthropic says its Life Sciences Verification Program uses safeguards built for legitimate professional research and development.

Mythos 5.1 can continue across long sequences of research, tool calls, and revisions, providing progress updates and recovering from some problems without restarting a task. It can work across large code repositories, write tests, and review proposed changes. Anthropic's Claude Security product now uses Mythos 5.1 to scan codebases and suggest patches for human review.

Mythos 5.1 also shares Fable 5.1's ability to complete multistep research and produce reports, spreadsheets, documents, and presentations, combining text and visual information from large collections of files.

Developers can select an effort level for each message. Higher effort gives the model more room to reason through difficult tasks, though it uses more tokens and can add latency. The default effort level is high.

Specifications

The one-million-token context window can hold a large collection of source code or research documents. Anthropic notes it does not guarantee perfect recall across every part of the input, so users should test the model with their own material before using it in production.

Availability in Nigeria and Africa

Claude Mythos 5.1 is currently unavailable to Nigerian organisations. Anthropic says it currently provides the model only to a set of approved US organisations. Although Project Glasswing previously expanded Mythos Preview to organisations in more than 15 countries, that earlier expansion does not confirm current Mythos 5.1 access in those countries.

Users in Nigeria can still access Claude Fable 5.1 through eligible paid plans or Anthropic's commercial API, but cannot select Mythos 5.1 through a regular Claude account. Anthropic says it plans to expand its trusted access programmes, though it has not published a date for Mythos 5.1 access in Nigeria or any other African country.

How to access Claude Mythos 5.1

Anthropic's Life Sciences Verification Program is an invite-only beta for advanced life science organisations, built with the US government, and has enrolled its first participants.

The Cyber Verification Program gives approved defenders access to selected Claude models with reduced cybersecurity safeguards. Anthropic says Mythos-class access will join this programme soon, though a full rollout date has not been announced — a milestone worth watching for security teams evaluating vetted access.

Project Glasswing is Anthropic's programme for organisations that build or protect important software systems. Approved partners use Mythos models for authorised defensive security work.

Organisations seeking approved access can contact their Anthropic, Amazon Web Services, or Google Cloud account team. Anthropic reviews the organisation and its intended use before granting access.

Supported platforms

Anthropic lists four supported platforms:

• Claude API, using the model ID claude-mythos-5-1
• Amazon Bedrock, using anthropic.claude-mythos-5-1
• Google Cloud, using claude-mythos-5-1
• Microsoft Foundry, subject to approved access

A normal account on any of these platforms does not provide automatic access; Anthropic must approve the organisation first.

Claude Mythos 5.1 vs Claude Fable 5.1

The two models match in general intelligence and specifications. Mythos can complete some specialist cybersecurity and biology work that Fable may stop or send to an Opus fallback model. Fable remains the version built for general users.

Mythos 5.1 keeps the standard $10 input and $50 output prices from Mythos 5, while its cache read price falls from $1 to $0.25 per million tokens. The new version also adds or improves:

• Long-running coding and multistep research
• Readable progress updates between tool calls
• Per-message effort controls and turn-specific system instructions
• Content provenance support for generated work

Organisations moving from Mythos 5 should review several API changes. Forced tool use is not supported, and earlier models cannot read Mythos 5.1 thinking blocks. Changes to earlier conversation content can also affect preserved reasoning, though Anthropic says Mythos does not run Fable 5.1's strict prefix integrity check.

Opus 5 costs half as much at standard API rates and runs with lower latency, making it suited to general reasoning and production work. Mythos 5.1 is built for approved organisations whose specialist research requires its trusted access safeguards.