- Anthropic released Claude Fable 5, the most powerful model in the company’s history to the public, on June 9, 2026, along with a limited Claude Mythos 5 for select cybersecurity partners.
- In the SWE-Bench Pro benchmark, the model achieved 80.3%, clearly ahead of GPT-5.5 (58.6%) and Gemini 3.1 Pro (54.2%).
- The API price is $10 per million input tokens and $50 per million output tokens, less than half the Mythos Preview rate.
Anthropic released on Tuesday, June 9, 2026, the first publicly available model from the new Mythos class. This is the same technology that the company has kept under lock and key since April out of fear of its cybersecurity capabilities. Fable 5 and the proprietary Mythos 5 use identical model weights. The difference comes down to the security layer: the public version blocks queries for cybersecurity, biology, chemistry, and model distillation, handing them over to the weaker Claude Opus 4.8. According to early data from Anthropic, over 95% of sessions are handled by Fable 5 alone, without triggering the failover mechanism.
Where did the Mythos class come from and what is Project Glasswing?
Anthropic first showed the Claude Mythos Preview in April. The model has proven to be extremely effective at finding security vulnerabilities in all major operating systems and browsers, even though no one designed it for this. The company concluded that such a powerful tool in the wrong hands could be used to attack critical infrastructure.
The answer was Project Glasswing. As part of the program, early access to Mythos Preview was granted to, among others: Amazon Web Services, Apple, Google, Cisco, Microsoft and JPMorgan Chase, which used the model to find and patch vulnerabilities in their own software. Last week, Anthropic expanded access to hundreds of organizations in 15 countries, again targeting entities managing critical infrastructure.
The Mythos class is the new, highest level in the hierarchy of Anthropic models, positioned above the current flagship Opus. Fable 5 is the version that the company has deemed safe for everyone.
Benchmarks: advantage over GPT-5.5 visible to the naked eye
The numbers are especially impressive in programming. On SWE-Bench Pro, a test that measures the ability of models to independently perform difficult software engineering tasks, Fable 5 scored 80.3%. For comparison, Claude Opus 4.8 achieves 69.2%, GPT-5.5 from OpenAI 58.6%, and Gemini 3.1 Pro from Google 54.2%. On the more difficult FrontierCode Diamond benchmark, the advantage is even greater: 29.3% compared to 13.4% for Opus 4.8 and only 5.7% for GPT-5.5.
Image analysis is also a strong point of the new model. Fable 5 can recreate the source code of a web app based solely on screenshots, and Pokémon FireRed went through the game from start to finish using only raw in-game images, without maps or navigation aids. Earlier versions of Claude needed extensive support tools for this.
The financial industry also tested the model. On the Hebbia financial benchmark, which tests reasoning at the level of an experienced analyst, Fable 5 achieved the highest result of all models, and trading company IMC reported that the model passed its trading analysis evaluations almost completely.
Pricing and Availability: Free access window until June 22
The API rate is $10 per million input tokens and $50 per million output tokens. This is a significant discount compared to the $25 and $125 that have been in effect for the closed Mythos Preview since April. A token is a basic unit of accounting in language models, simply put, a fragment of a word. However, the new model is twice as expensive as Claude Opus 4.8, the company’s current flagship.
What does the premiere mean for the crypto market
For the cryptocurrency industry, this is a double-edged sword. Mythos models can quickly find subtle vulnerabilities in smart contracts, node software and exchange infrastructure. This shortens the time between the appearance of a vulnerability and its exploitation, which raises the stakes in an industry that regularly loses hundreds of millions of dollars to exploits. On the other hand, these same tools speed up audits and code patching, giving an advantage to the teams that implement them first.
The premiere also has a purely market dimension. Anthropic is capitalizing on growing investor interest ahead of a potentially gigantic IPO that could take place later this year. The company recently raised $65 billion in a financing round at a valuation of $965 billion and filed a confidential IPO application. Annual revenues reached approximately USD 47 billion in May, up from approximately USD 10 billion a year earlier. Analysts treat a debut above a trillion dollars as a base scenario, although the mood around the valuations of AI companies remains variable, and part of the market openly talks about the risk of a bubble.
