# Meta’s Muse Spark Open model now on Fireworks US serverless API

> The 816‑billion‑parameter open‑weights model is live day‑0, offering 42 B active parameters per token and a 1 M‑token context for US‑based legal, finance and coding tasks.

Oossa · 2026-10-20 · https://oossa.com/en/meta-s-muse-spark-open-model-now-on-fireworks-us-serverless-api

Fireworks AI announced that Muse Spark Open, Meta’s open‑weights model, is available today on its US‑only serverless API. The model uses a mixture‑of‑experts architecture with 816 B total parameters but only 42 B active per token, and it can handle up to a 1 M‑token context window. Fireworks runs it with zero data retention by default and offers a serverless‑to‑reserved‑throughput upgrade path, plus LoRA fine‑tuning.

## The facts

- 816 B total parameters, 42 B active per token, 1 M‑token context window
- Ranks #1 on Vals AI’s Harvey’s Legal Agent Benchmark with a 25.4 % pass rate, about four times Claude Opus 5

## Why it matters

Enterprises in regulated fields can now run a high‑performing US‑trained model locally without data retention, reducing compliance risk and cost.

## Sources & references

1. [Muse Spark Open on Fireworks: Meta’s open weights wodel, available on US serverless](https://fireworks.ai/blog/meta-muse-spark-open) – Fireworks AI

Last updated: 2026-10-20