Crouton Alpha reasons like Meta's Muse models, but it counts tokens like none of them. Our lead: a new model from Meta.
By Stealth Models.
Facts as text
| Fact | Value |
|---|---|
| First seen | 3 Nov 2026 |
| Revealed | not yet |
| Developer | unknown, likely Meta |
| Context | 1,048,576 tokens |
| Input | Text, images, audio, video, files |
| Tool calls | Yes |
| Speed in our tests | about 175 tok/s |
| Access | Free preview on OpenRouter and OpenCode, blocked in Europe; prompts may be kept by the provider, not used for training |
Weights
So far Crouton Alpha is only a hosted preview. If it turns out to be an open model, for example a new open Muse like Muse Spark Open, we'll put GGUF files for llama.cpp here, as we did for GLM-5.3-Flash and MiniMax M3.1 Flash.
StealthMark leaderboard
Text and images: CC BY 4.0, stealthmodels.com.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support