On 9 November, two providers unveiled a new model, from two sides of the market. Anthropic released Claude Fable 5.2, faster and cheaper than the previous version. Meta launched Muse Ember, a small model for large volumes of routine work. What does that mean for your costs and your privacy?
Claude Fable 5.2: faster, lower price per token
According to Anthropic's announcement, Fable 5.2 makes its answers up to 30% faster than Fable 5.1. Via the API the model costs $8 per million input tokens and $40 per million output tokens, and Anthropic explicitly describes that as a lower price. Anyone choosing the fast mode pays $16 and $80 and, according to Anthropic, gets text back up to 2.5 times as quickly. The model is available in Claude.ai, Claude Code and Cowork, and on Amazon Web Services, Google Cloud and Microsoft Foundry.
At the same time, Claude Mythos 5.2 appeared. It is the same model with fewer restrictions, and is only intended for verified security specialists and life sciences researchers. Claude Enterprise customers can use Mythos 5.2 via Claude Security.
What does this mean for you? If you use Claude via the API, your costs drop for the same usage. Mind data retention: according to the same page, Anthropic stores inputs by default for 30 days for safety checks. Only Enterprise customers with Enterprise Frontier Safeguards get zero data retention, and that is still in a closed beta. This is not new policy, but it is something to weigh if you work with personal data. For how EU hosting and training work, read the Claude page and the GDPR checks for software development and government. If you work in cybersecurity, the verification programme for Mythos 5.2 is the route.
Meta Muse Ember: small model for many tasks
Meta presented Muse Ember as its smallest multimodal model. According to Meta's model page, it costs $0.10 per million input tokens and $0.50 per million output tokens via the Meta Model API, up to 200,000 tokens of context. Beyond that it becomes $0.30 and $1.50. The context window is 1 million tokens, and the model reads text, images, audio and video. It is also available via Google Cloud Marketplace and OpenRouter. SiliconANGLE reports that Meta positions the model for agents that carry out many small tasks, such as summarising documents, extracting data and code subtasks. According to Meta, it matches the performance of Claude Haiku 5.5 for about a third of the price. That is a claim by Meta itself, not yet independently tested.
What does this mean for you? For simple, repetitive work, such as extracting data from invoices or sorting customer enquiries, a small model can make a considerable difference in costs. Meta's model page says nothing about where your data is processed or how long it is kept. Until that is clear, do not send personal data to it. This is different from the open Llama models you can host yourself: compare those on the Llama page and in the GDPR check for customer service. A comparison of both families is at Claude vs Llama.
In brief
- Claude Fable 5.2 is up to 30% faster than Fable 5.1 and costs $8 and $40 per million tokens via the API.
- Anthropic stores API inputs for 30 days by default; zero retention only in an Enterprise beta.
- Meta Muse Ember costs $0.10 and $0.50 per million tokens; Meta gives no information on processing location.
- All models and privacy scores can be found under models and safety.