Meta (Muse)
External APIs
Meta serves Muse Wonder at $4 and $20 per million. Muse Wonder Full has no per-token price and is obtained by subscription.
Key points : Muse Wonder · No training on standard services · Zero data retention available.
Models served : the range opens with Muse Wonder, released on 12 October 2026, at $4 input and $20 output per million, $0.20 cache, for a one-million-token context. Muse Wonder Full, the higher model, is not sold per token and comes with a Muse Code subscription at $50 or $200 per month. Muse Spark Open is its open-weight variant.
Where the data goes
Meta writes that it does not use content from standard services to train its models. That content may be processed to detect and address usage policy violations, and to develop safety systems. Exchanges are encrypted in transit and subject to access controls.
Retention and training
A zero data retention option exists for qualified accounts. Prompts and responses are then processed in real time and not stored after the response. It is enabled at organisation level, never from the web interface. Even then, a flagged request may be kept for up to two years.
When to use it from a sovereign platform
The written commitment not to train on standard services puts Meta ahead of several competitors. Zero retention is worth requesting for professional use. Since the higher model is sold by subscription, it escapes the per-token arithmetic of our comparisons.