Anthropic unveils Claude Fable 5, a cautious successor to Mythos
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Anthropic Unveils Claude Fable 5, a Cautious Successor to Mythos
Two months ago, Anthropic made waves with its AI model, Mythos, capable of autonomously detecting and exploiting cybersecurity vulnerabilities. However, due to the sensitivity of its capabilities, the company decided not to make it publicly accessible. Today, the situation evolves with the launch of Claude Fable 5.
Claude Fable 5: A Public Version of Mythos
Claude Fable 5 is essentially a version of Mythos 5, but under a new name. This model retains the power of its predecessor while introducing usage restrictions. On the SWE-Bench Pro benchmark test, Claude Fable 5 achieves an impressive score of 80.3%, surpassing other models like Claude Opus 4.8 at 69.2%, GPT-5.5 at 58.6%, and Gemini 3.1 Pro at 54.2%.
Performance and Restrictions
Claude Fable 5 stands out particularly on complex tasks. For instance, on the FrontierCode Diamond benchmark, it reaches 29.3%, compared to 13.4% for Opus 4.8. Additionally, it successfully completed Pokémon Fire Red solely from screenshots, without assistance, a feat that previous versions of Claude could not achieve without additional tools.
However, Claude Fable 5 incorporates monitoring systems that analyze requests in real time. If a request pertains to sensitive areas such as offensive cybersecurity, biology, chemistry, or the reproduction of AI models, it is automatically redirected to Claude Opus 4.8. Anthropic claims that this filtering affects less than 5% of conversations.
Availability and Pricing
Claude Fable 5 is accessible via the Claude API, Claude Code, as well as on AWS, Google Cloud, and Microsoft Foundry. Pro, Max, Team, and Enterprise subscribers can use it for free until June 22, after which usage will be charged based on dedicated credits. For developers, the API is priced at $10 per million input tokens and $50 per million output tokens, making it one of the most expensive models from Anthropic.
Meanwhile, Mythos 5 remains reserved for a select group of organizations as part of the Project Glasswing program.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.