What is Claude Fable 5? How much does it cost? View all Mythos level super model features at once

Anthropic Launches Claude Fable 5 and Mythos 5, Completely New Mythos-Level AI Models. This article introduces what Fable 5 is, how it is priced, and its main features and highlights.

What is Claude Fable 5? A Lazy Summary for You

AI giant Anthropic surprised everyone last night (6/9) with the release of two new AI models, Claude Fable 5 and Claude Mythos 5. Officially, both models are Mythos level, with capabilities higher than Opus, designed specifically for performing complex reasoning and long-process agent (Agent) tasks.

This is Anthropic’s first time opening its most powerful top-tier models to the general public. Previously, the Mythos preview version (Claude Mythos Preview) launched in April was only privately available to a select few partners due to its strong ability to identify security vulnerabilities online. The Claude Fable 5 released this time is a public version with added security protections.

Claude Fable 5 Pricing: Twice the Cost of Opus 4.8

Currently, the pricing for Claude Fable 5 and Claude Mythos 5 is $10 per million input tokens and $50 per million output tokens, which is twice the price of the previous flagship model Opus 4.8.

For general users and enterprise clients, Anthropic has planned different phased access schedules:

• API and Enterprise Metered Plans: Starting June 9, 2026, global developers can access Claude via API (model ID: claude-fable-5) and platforms like AWS, Google Cloud Vertex AI directly.

General subscription users (Pro, Max, Team plans):

• First phase (June 9 to June 22): Subscribers can experience Claude Fable 5 for free on the platform, with no extra charge.
• Second phase (from June 23): Due to initial global computing resource constraints, Anthropic will temporarily remove Fable 5 from standard subscription plans. Users will need to purchase usage credits for continued use.
• Future plans: Diane Penn, product manager at Anthropic, told Wired that current protections and quota limits are to ensure user access to value while maintaining safety. Once server infrastructure is expanded, the goal is to quickly reintegrate Fable 5 into standard subscription plans.

Who Can Use Claude Mythos 5?

As for Claude Mythos 5, which lacks a safety classifier, it remains strictly controlled at this stage.

Currently, only the US government partnership project Glasswing will upgrade its previous preview version. Subsequently, access will be gradually opened through a review-based “Trusted Access Program” to vetted cybersecurity organizations and biomedical research institutions.

Claude Fable 5 Features and Use Cases

Based on Anthropic’s testing data and feedback from third-party companies, Claude Fable 5 demonstrates significant breakthroughs in software engineering, knowledge work, and visual analysis, with five core features:

Major Boost in Software Engineering Efficiency

Anthropic’s official data shows that Fable 5 has a clear advantage in handling complex and time-consuming engineering tasks.

Stripe, a corporate software company, reported early testing that Fable 5 shortened months of engineering optimization work to just a few days.

When working on a Ruby project with up to 50 million lines of code, the model completed the entire codebase migration in just one day. Doing this manually with an engineering team would typically take over two months.

Image source: AnthropicClaude Fable 5’s advantage in handling complex, time-consuming engineering tasks

Advanced Knowledge Work and Reasoning Capabilities

Fable 5 also excels in complex data analysis and high-level logical reasoning.

In Hebbia’s advanced financial reasoning benchmark tests, Fable 5 achieved the highest scores among existing models, especially in document reasoning, chart analysis, and problem-solving.

Analysis firm Hex stated that Fable 5 is the first model to score 90% on its core analytical benchmarks, demonstrating excellent judgment and attention to detail in challenging long-process analysis tasks.

Cutting-Edge Visual Analysis Technology

Claude Fable 5 sets new standards in visual tasks, capable of extracting precise data from complex scientific charts and even reconstructing original code from screenshots of web applications.

Anthropic claims that previous Claude models, even with auxiliary tools, struggled with complex visual game tasks. But Fable 5, with only basic visual analysis tools and no map or navigation aids, successfully completed the classic Nintendo game Pokémon FireRed.

Image source: AnthropicClaude Fable 5 can beat Pokémon FireRed

Long Text and Autonomous Memory

Claude Fable 5 and Mythos 5 support a context window of 1 million (1M) tokens, with each request capable of handling up to 128k (128k) tokens of output.

Both models enforce an “Adaptive Thinking” mode in their API operation.

Ethan Mollick, a researcher and AI scientist at the University of Pennsylvania, tested and posted on Substack that Fable 5 significantly outperforms any public model he has used, capable of continuously executing multi-page complex specifications for over ten hours.

In tests, Fable 5 generated complete, dynamic game templates like Snake and Underground Lamp with just a single initial prompt.

Image source: AnthropicClaude Fable 5 can automatically generate simple game templates

Scientific R&D and Hypothesis Generation

Anthropic’s internal protein design experts note that using Claude Mythos 5 with bioinformatics tools can accelerate drug design processes by about ten times without human assistance, achieving results comparable to professional scientists.

In blind tests, scientists rated the molecular biology hypotheses proposed by Mythos 5 with an 80% satisfaction rate. One hypothesis regarding E. coli proteins was later confirmed by an independent external lab.

Image source: AnthropicClaude Mythos 5 in scientific applications

Mythos 5, Fable 5, Opus 4.8 Comparison

According to Anthropic’s official technical documentation, the following table summarizes the specifications differences between the newly released Mythos models and the current flagship Claude Opus 4.8 (mobile users can swipe horizontally):

| Model | API Model ID | Price per Million Tokens | Security Measures | Open To | | --- | --- | --- | --- | --- | | Claude Fable 5 | claude-fable-5 | $10 / $50 | Yes, built-in high-risk domain classifier | Global API users, enterprise users, and subscribers | | Claude Mythos 5 | claude-mythos-5 | $10 / $50 | No, no network security or biological restrictions | Glasswing project partners and select researchers | | Claude Opus 4.8 | claude-opus-4.8 | $5 / $25 | Standard security | General public worldwide |

Claude Fable 5 Security Protections

According to Wired, because Mythos-level models are so powerful, their web security defenses and vulnerability detection capabilities could be exploited maliciously, leading to serious cyberattack threats.

To balance safety and release speed, Anthropic has implemented special protective measures.

Automatic Downgrade (Fallback) Mechanism

Claude Fable 5 has an independent AI safety classifier. When user prompts touch on three high-risk areas, the system will refuse to respond with Fable 5 and automatically redirect and downgrade the request to Claude Opus 4.8:

• Cybersecurity: Including exploits, attack planning, or defense evasion.
• Biological and Chemical: Involving high-risk biological research or potential chemical weapons.
• Model Distillation: Detecting if users attempt to leverage Fable 5 outputs to train competing models.

Anthropic emphasizes that this classifier’s filtering criteria are currently conservative, occasionally misjudging harmless requests. However, initial data shows over 95% of Fable 5 interactions do not trigger downgrades, allowing users to experience full Fable 5 performance.

When a request is downgraded, users are notified, and if intercepted before generating any text, the request will not be billed.

30-Day Data Retention to Prevent Jailbreaks

TechCrunch reports that to prevent sophisticated security vulnerabilities and jailbreak attacks across multiple requests, Anthropic enforces a mandatory 30-day data retention policy for Fable 5 and Mythos 5.

Regardless of whether enterprise users previously signed zero data retention agreements, all traffic to and from these models is recorded and stored for 30 days.

Anthropic commits that this data will never be used for training new models, and access logs for internal personnel will be deleted after expiration.

Further Reading:
Anthropic CEO warns: Mythos has identified tens of thousands of software vulnerabilities, with only 6-12 months left to fix them

Ultraman preview: OpenAI’s new security model GPT 5.5-Cyber to debut in a few days, competing with Claude Mythos