Mistral Large 4 vs Muse Spark Open: two giant open-weights models
Mistral Large 4 and Muse Spark Open were both released in October 2026. Compare size, input types, memory needs and the license conditions of each.
Verdict. Both were released in October 2026 and neither is a local model for most people. Mistral Large 4 (1.05T parameters, 52B active) reads text and images in more than 160 languages and is tuned for coding and cybersecurity; its license is generous except that hosting it as a paid service for others needs a separate license. Muse Spark Open (816B, 42B active) also understands video and audio, is gated on Hugging Face and asks everyone to credit 'Built with Muse Spark Open'.
Specifications
| Specifications | Mistral Large 4 | Muse Spark Open |
|---|---|---|
| Developer | Mistral AI | Meta |
| Released | October 29, 2026 | October 15, 2026 |
| Total parameters | 1.05T (Large 4) | 816B (Spark Open) |
| Active parameters | 52B | 42B |
| Architecture | Mixture of experts | Mixture of experts |
| Context window | 1M tokens | 1M tokens |
| Input | Text, Image | Text, Image, Video, Audio |
| License | Mistral Open-Weight (Conditional) | Muse Model License (Conditional) |
| Memory for the smallest variant (4-bit) | 671 GB (Large 4 NVFP4) | 563 GB (Spark Open) |
Performance claims come from the developer and have not been independently verified.
Pick Mistral Large 4 if
- You need data sovereignty or an on-premises deployment, especially in Europe, with a model trained from scratch there.
- You want strong European-language coverage: Mistral lists more than 160 languages.
- Your work is coding agents or defensive security.
- You do not plan to resell hosted access to the model.
Pick Muse Spark Open if
- You need audio and video understanding along with text and images.
- You want a Meta model you can host yourself for coding and agent workloads with tool use.
- You can accept the gated download and show the 'Built with Muse Spark Open' credit.
- You want a smaller sibling under different terms: Muse Glimmer is a 30B dense model under Apache 2.0.