One Model, Two Versions

Both Claude Fable 5 (the public release) and Claude Mythos 5 (the restricted release) represent Anthropic's new "Mythos" class — its most capable AI tier ever . The core architecture is identical . What changes between them is not performance, but permission: Fable 5 puts hard safety limits on what you can ask, while Mythos 5 opens the door to sensitive domains like cybersecurity vulnerability discovery and advanced life sciences research .

Core Capabilities Shared by Both Models

The models deliver state-of-the-art results across nearly all published benchmarks . Here are the headline capabilities they share:

• 1M token context window by default, with up to 128,000 output tokens per request .
• Always-on adaptive thinking, where the model's reasoning effort is configurable but cannot be disabled .
• Autonomous agentic work: The models can plan across stages, delegate to sub-agents, and self-verify their work over multi-day tasks .
• Software engineering at scale: Stripe used Fable 5 to perform a codebase-wide migration across 50 million lines of Ruby in a single day—a task that would have taken a team over two months .
• Vision and reasoning: The model reaches a new state of the art on visual tasks, including rebuilding web application source code from screenshots alone .
• Drug discovery and science (Mythos 5): Mythos 5 accelerated aspects of drug design by roughly 10x, and in blind studies, 9 of 14 protein targets yielded strong drug-design candidates. It is also the first Claude model to consistently produce novel molecular biology hypotheses .

Safety Trade-Offs: Where the Two Models Diverge

The safety architecture is the single biggest difference between the two models.

Claude Fable 5: Conservative Safety Classifiers

Fable 5 runs safety classifiers on both incoming requests and during response generation . When a classifier triggers, the API returns a stop_reason: "refusal" with an HTTP 200 status code—not an error—and identifies which classifier was activated . For queries in high-risk areas — specifically cybersecurity, biology, chemistry, and model distillation — Fable 5 does not simply refuse. It falls back to Claude Opus 4.8, Anthropic's prior-generation model, to provide a safer answer .

Anthropic states these safeguards fire in less than 5% of sessions on average, but acknowledges they are tuned conservatively and will occasionally flag harmless requests .

Claude Mythos 5: Safeguards Lifted in Specific Areas

Claude Mythos 5 removes those classifiers in certain domains, granting what Anthropic describes as "the strongest cybersecurity capabilities of any model in the world" to its authorized users . This model is not sold on the open market. It is made available exclusively through Project Glasswing, a partnership with the U.S. government, for cyberdefenders and critical infrastructure providers . Anthropic has stated it intends to expand access to trusted partners over time .

Pricing: Identical for Both Models

Despite the enormous difference in access and risk profile, both models are priced identically :

• Input tokens: $10 per million tokens
• Output tokens: $50 per million tokens

This pricing is less than half the cost of the earlier Claude Mythos Preview . On Claude.ai subscription plans, Fable 5 usage counts as 2x usage, and it will be free on Pro, Max, Team, and seat-based Enterprise plans from June 9–22, 2026, after which it transitions to a usage-credit system .

Data Retention: A Key Limitation

Both Fable 5 and Mythos 5 have a 30-day data retention policy. They are designated as Covered Models and are explicitly not available under zero data retention terms . Organizations with strict data-residency or deletion requirements need to account for this before deploying either model.

Access: Who Can Use Which Model

The availability gap is absolute for now:

• Claude Fable 5 is generally available to enterprise customers, developers, and individual subscribers through the Claude API, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry .
• Claude Mythos 5 is available only through Project Glasswing. Access is limited, handled directly by Anthropic, AWS, and GCP account teams, and currently reserved for U.S. government cyberdefense partners and critical infrastructure providers .

The Bigger Picture: Why This Split Matters

This dual launch is a new model for frontier AI deployment: ship the most powerful system you can build, but gate its most dangerous capabilities behind a government-trusted access program. For the vast majority of developers and enterprises, Fable 5 is the most capable model Anthropic has ever made publicly available, with the price, performance, and context window to power genuinely new use cases in software engineering and long-horizon autonomous work . For security researchers and biodefense teams inside Project Glasswing, Mythos 5 unlocks a fundamentally different tier of capability for tasks the public version will not touch.

One Model, Two Versions

Both Claude Fable 5 (the public release) and Claude Mythos 5 (the restricted release) represent Anthropic's new "Mythos" class — its most capable AI tier ever . The core architecture is identical . What changes between them is not performance, but permission: Fable 5 puts hard safety limits on what you can ask, while Mythos 5 opens the door to sensitive domains like cybersecurity vulnerability discovery and advanced life sciences research .

Core Capabilities Shared by Both Models

The models deliver state-of-the-art results across nearly all published benchmarks . Here are the headline capabilities they share:

• 1M token context window by default, with up to 128,000 output tokens per request .
• Always-on adaptive thinking, where the model's reasoning effort is configurable but cannot be disabled .
• Autonomous agentic work: The models can plan across stages, delegate to sub-agents, and self-verify their work over multi-day tasks .
• Software engineering at scale: Stripe used Fable 5 to perform a codebase-wide migration across 50 million lines of Ruby in a single day—a task that would have taken a team over two months .
• Vision and reasoning: The model reaches a new state of the art on visual tasks, including rebuilding web application source code from screenshots alone .
• Drug discovery and science (Mythos 5): Mythos 5 accelerated aspects of drug design by roughly 10x, and in blind studies, 9 of 14 protein targets yielded strong drug-design candidates. It is also the first Claude model to consistently produce novel molecular biology hypotheses .

Safety Trade-Offs: Where the Two Models Diverge

The safety architecture is the single biggest difference between the two models.

Claude Fable 5: Conservative Safety Classifiers

Fable 5 runs safety classifiers on both incoming requests and during response generation . When a classifier triggers, the API returns a stop_reason: "refusal" with an HTTP 200 status code—not an error—and identifies which classifier was activated . For queries in high-risk areas — specifically cybersecurity, biology, chemistry, and model distillation — Fable 5 does not simply refuse. It falls back to Claude Opus 4.8, Anthropic's prior-generation model, to provide a safer answer .

Anthropic states these safeguards fire in less than 5% of sessions on average, but acknowledges they are tuned conservatively and will occasionally flag harmless requests .

Claude Mythos 5: Safeguards Lifted in Specific Areas

Claude Mythos 5 removes those classifiers in certain domains, granting what Anthropic describes as "the strongest cybersecurity capabilities of any model in the world" to its authorized users . This model is not sold on the open market. It is made available exclusively through Project Glasswing, a partnership with the U.S. government, for cyberdefenders and critical infrastructure providers . Anthropic has stated it intends to expand access to trusted partners over time .

Pricing: Identical for Both Models

Despite the enormous difference in access and risk profile, both models are priced identically :

• Input tokens: $10 per million tokens
• Output tokens: $50 per million tokens

This pricing is less than half the cost of the earlier Claude Mythos Preview . On Claude.ai subscription plans, Fable 5 usage counts as 2x usage, and it will be free on Pro, Max, Team, and seat-based Enterprise plans from June 9–22, 2026, after which it transitions to a usage-credit system .

Data Retention: A Key Limitation

Both Fable 5 and Mythos 5 have a 30-day data retention policy. They are designated as Covered Models and are explicitly not available under zero data retention terms . Organizations with strict data-residency or deletion requirements need to account for this before deploying either model.

Access: Who Can Use Which Model

The availability gap is absolute for now:

• Claude Fable 5 is generally available to enterprise customers, developers, and individual subscribers through the Claude API, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry .
• Claude Mythos 5 is available only through Project Glasswing. Access is limited, handled directly by Anthropic, AWS, and GCP account teams, and currently reserved for U.S. government cyberdefense partners and critical infrastructure providers .

The Bigger Picture: Why This Split Matters

This dual launch is a new model for frontier AI deployment: ship the most powerful system you can build, but gate its most dangerous capabilities behind a government-trusted access program. For the vast majority of developers and enterprises, Fable 5 is the most capable model Anthropic has ever made publicly available, with the price, performance, and context window to power genuinely new use cases in software engineering and long-horizon autonomous work . For security researchers and biodefense teams inside Project Glasswing, Mythos 5 unlocks a fundamentally different tier of capability for tasks the public version will not touch.