# Baseten — Muse Spark Open, an open-weight model for coding agents

- Company: Baseten (baseten.co)
- Announced: 2026-10-15T17:27:04+00:00
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.baseten.co/blog/muse-spark-open-launch
- Record: https://forck.live/items/23008-muse-spark-open-an-open-weight-model-for-coding-agents
- Subject: Baseten
- Open weights: yes
- Models affected: Muse Spark Open
- Context window: 1M token context window

Meta released the weights for Muse Spark Open, an 816B-parameter open-weight omnimodal MoE model co-trained with Muse Code for coding agent workflows. The model has a 1M token context window, uses a compressed MoE block and hybrid global/local attention to reduce inter-GPU traffic and KV cache memory. It is available on Baseten Model APIs and Dedicated Inference served in NVFP4.

## Evidence

Verbatim from https://www.baseten.co/blog/muse-spark-open-launch:

> Today, Meta released the weights for Muse Spark Open, an 816B-parameter open-weight omnimodal MoE model co-trained with Muse Code for coding agent workflows.

---

Record: https://forck.live/items/23008-muse-spark-open-an-open-weight-model-for-coding-agents
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md