Meta Muse Spark 1.3 🧠 Model Freemium

What it is

Meta's language models used to mean Llama. In April 2026 Meta Superintelligence Labs replaced Llama in its own products with Muse Spark, a proprietary model; Muse Spark 1.3 followed on 2 September 2026. On 10 August 2026 Meta returned to open weights with Muse Glimmer, a 30B multimodal model under Apache 2.0. The last Llama release is still Llama 4 (Scout and Maverick, April 2025), whose weights remain downloadable.

Key features

• Muse Spark 1.3 sustains longer tool-based work and asks clarifying questions, per Meta, using about 20% fewer tool calls than 1.2
• Muse Glimmer runs locally in under 20 GB at 4-bit, on one consumer GPU or an Apple-silicon Mac
• Glimmer ships quantised variants and works with llama.cpp, MLX, vLLM, SGLang, Ollama and LM Studio
• Llama 4 Scout offers up to a 10M-token context window; Maverick 1M

Where you can use it

Muse Spark runs in Meta AI across Meta's apps and on the Meta Model API at dev.meta.ai. Muse Glimmer and Llama 4 weights are on Hugging Face and through hosts such as Together AI and Fireworks AI. None of the apps in our directory document Muse or Llama as a selectable model, so the app list here is empty; local runners like Ollama are the usual route.

Pricing and rights

Muse Spark 1.3 API: $1.25 input / $4.25 output per million tokens. Meta AI is free to use. Muse Glimmer is Apache 2.0, so commercial use is allowed; Llama 4 uses the Llama 4 Community License with its own conditions.

Who it is for

Creators and developers who want a private, local writing model, and teams building on Meta's API.

Verdict

Meta's open story is now Muse Glimmer rather than Llama, and it is a practical local model. The flagship Muse Spark is closed and aimed at coding agents, so for writing quality most creators will still pick another family.