Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9, 2026. These two models open a new tier called Mythos, positioned above the Opus class in raw capability. For the first time, a model of this level reaches the general public through Fable 5, while Mythos 5 stays reserved for approved organisations.
This article explains what these models are, what separates them, their measured performance, and what the release changes in practice for a small business that uses AI in its marketing operations.
In short. Fable 5 and Mythos 5 run on the same model. Fable 5 is available to everyone with its guardrails. Mythos 5 has some protections lifted and stays reserved for approved organisations. API pricing: $10 input and $50 output per million tokens.
Fable 5 and Mythos 5: one model, two levels of guardrails
Fable 5 and Mythos 5 run on the exact same underlying model. The only difference is the safety layer. Anthropic frames it through the etymology of the two names: Fable comes from the Latin fabula, "that which is told", close to the Greek mythos. The safeguards are what set the two versions apart.
Claude Fable 5 is the generally available version, with its protections active. Claude Mythos 5 is the same engine with some protections lifted, accessible only to cybersecurity professionals, infrastructure providers and a selection of biomedical researchers.
In practice, Fable 5 ships with classifiers that detect sensitive requests (offensive cyber, biology, chemistry) and route flagged queries to Claude Opus 4.8. Anthropic states that more than 95% of Fable sessions trigger no fallback at all. When a fallback happens, the response comes from Opus 4.8, itself a highly capable model. For marketing, sales or productivity work, these guardrails stay invisible.
Claude Fable 5 measured performance
Published benchmarks place Fable 5 at the top of several reference evaluations, especially on code. The table below sums up the gaps on SWE-bench Pro, which measures the resolution of real GitHub issues.
| Model | SWE-bench Pro | FrontierCode (Diamond) |
|---|---|---|
| Claude Fable 5 | 80.3% | 29.3% |
| Claude Opus 4.8 | 69.2% | 13.4% |
| GPT-5.5 | 58.6% | 5.7% |
| Gemini 3.1 Pro | 54.2% | — |
On SWE-bench Pro, Fable 5 reaches 80.3%, 11 points ahead of Opus 4.8, more than 20 points ahead of GPT-5.5 and Gemini 3.1 Pro. On SWE-bench Verified, the model climbs to 95.0%. On Cognition's FrontierCode evaluation, in its hardest Diamond split, it scores 29.3%, more than double Opus 4.8.
Vision is the second area of marked progress. Anthropic describes Fable 5 as a reference model for visual tasks: extracting numbers from scientific figures, rebuilding the source code of a web app from a screenshot. As a demonstration, the model finished the game Pokémon FireRed using vision alone.
A third axis concerns long-context memory. The model keeps its coherence across millions of tokens. On the game Slay the Spire, its persistent memory tripled its results compared with Opus 4.8.
Real enterprise case. Stripe migrated a 50-million-line Ruby codebase in a single day, against two months of manual work.
Pricing and access
Claude Fable 5 costs 10 dollars per million input tokens and 50 dollars per million output tokens through the API. Anthropic notes that this is less than half the price of the preliminary Mythos Preview release from April 2026. The model API identifier is claude-fable-5.
| Item | Detail |
|---|---|
| Input price (API) | $10 / million tokens |
| Output price (API) | $50 / million tokens |
| Included in Pro, Max, Team, Enterprise | June 9 to 22, 2026, no extra cost |
| From June 23, 2026 | usage consumes credits |
| Data retention (Mythos class) | 30 days, no training |
API access and consumption-based Enterprise access have been available since launch.
Mythos 5 stays limited to Project Glasswing partners, focused on cybersecurity and infrastructure. A gradual opening to biomedical researchers is planned, with cyber protections kept in place.
For Mythos-class model traffic, Anthropic applies a 30-day data retention, with no use for training new models. All human access is logged, and data is deleted after that period in most cases.
What the release changes for a small business
A more capable model does not automatically transform a business. Its value depends on real use cases. Here are the areas where Fable 5 brings a concrete gain for a mid-sized organisation.
Content production gains reliability. Adherence to tone, format and structure improves with each model generation. A precise brief produces a first draft closer to the final result, which cuts editing time. This applies to articles, product descriptions, email sequences and service pages.
Data analysis becomes accessible without advanced technical skills. The model's vision capability lets it read a dashboard, a chart or an export as an image and draw a summary from it. A manager can submit a screenshot of their advertising stats and get a reading of the key points.
Development and automation reach a new level. The coding scores mean a model can now handle technical tasks once reserved for a developer: building an automation script, fixing a bug, connecting two tools through their APIs. For a small business with no technical team, this widens the range of feasible projects.
Large document handling gets simpler. Keeping coherence across very long contexts allows the model to process a full case file, a set of contracts or a knowledge base in a single pass, without artificially splitting the work.
How to integrate these models into your business
Having a capable model is not enough. The value comes from its integration into specific processes. An agency that treats AI as a cross-functional method builds AI agents tailored to each need: lead qualification, quote generation, handling recurring requests.
Model choice depends on the task. For high-volume, low-complexity operations, an economical model like Haiku stays relevant. For strategic analysis, code generation or advanced reasoning, Fable 5 justifies its cost. This use-case-based selection avoids paying for a premium model on tasks that do not need it.
Guidance matters as much as the tool. A small business that wants to benefit from these models needs to frame its use cases, train its teams and secure its data handling. That is the purpose of structured AI consulting, which starts from business needs before choosing the technology.
Frequently asked questions
What is the difference between Claude Fable 5 and Claude Mythos 5?
The two models are technically identical. Fable 5 has active safety guardrails and is available to everyone. Mythos 5 has some protections lifted and stays reserved for approved organisations in cybersecurity, infrastructure and biomedical research.
How much does Claude Fable 5 cost?
Through the API, the price is 10 dollars per million input tokens and 50 dollars per million output tokens. From June 9 to 22, 2026, the model is included at no extra cost in the Pro, Max, Team and Enterprise plans, then its usage consumes credits from June 23.
Is Claude Fable 5 better than GPT-5.5?
On the published coding benchmarks, Fable 5 leads GPT-5.5 by more than 20 points on SWE-bench Pro (80.3% versus 58.6%). On other task types, model choice depends on the use case and deserves a test on your own data.
Is Claude Fable 5 suitable for a small business with no technical skills?
Yes. The model works through a chat interface for writing, analysis or summarisation, with no technical knowledge. For advanced uses such as automation or API integration, guidance helps frame the projects.
Is my data used to train the model?
For Mythos-class model traffic, Anthropic applies a 30-day retention and does not use the data to train new models. Conditions vary by access type and plan, which is worth checking case by case.