---
type: "post"
id: "muse-spark-open-on-fireworks-meta-s-open-316301"
slug: "muse-spark-open-on-fireworks-meta-s-open-316301"
title: "Muse Spark Open on Fireworks: Meta’s Open Weights Model, Available on US Serverless"
description: "**Thesis:** A model’s open weights create options, but the serving route determines which teams can put those options into production."
retrieval_nugget: "**Thesis:** A model’s open weights create options, but the serving route determines which teams can put those options into production."
published_at: "2026-10-21"
updated_at: "2026-10-21"
record_date: "2026-10-21"
date_kind: "published_at"
topics: ["open-models","inference"]
source_urls: ["https://fireworks.ai/blog/meta-muse-spark-open"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:muse-spark-open-on-fireworks-meta-s-open-316301"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:cf0fb90f3593119537ba9f6da50e4839bc334e07b948ae50c8504bf57a0026c3","reviewed_at":"2026-10-21T08:05:08.924Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"A model’s open weights create options, but the serving route determines which teams can put those options into production.","observed_facts":[{"text":"Fireworks says Muse Spark Open is available on its serverless API using US-only capacity, with no US premium and Zero Data Retention by default.","source_urls":["https://fireworks.ai/blog/meta-muse-spark-open"]},{"text":"The source describes an 816-billion-parameter model with 42 billion active parameters per token and a one-million-token context window.","source_urls":["https://fireworks.ai/blog/meta-muse-spark-open"]},{"text":"Fireworks offers reserved throughput for production workloads and serverless LoRA training.","source_urls":["https://fireworks.ai/blog/meta-muse-spark-open"]}],"mechanism":"A distinct hosting route packages access to open weights with stated geographic capacity, data-retention defaults, and options for production throughput and model adaptation.","why_now":"Fireworks’ announcement adds a provider-specific serving option after earlier reports of the model’s availability on other platforms.","implications":["Teams can compare another hosting route against their geographic, data-handling, and workload requirements.","Reported benchmark results are not interchangeable because the announcement describes different evaluators and harnesses."],"evidence_boundary":"The evidence is Fireworks’ company announcement. Its performance figures are vendor-reported or benchmark-specific; the source says some third-party results were measured on Meta’s API release. It does not establish comparative performance on a reader’s workload or independently verify service terms.","watch_conditions":["Watch for independent, same-workload tests across providers and confirmation of current pricing and operational terms for reserved throughput and LoRA training."],"related_records":[],"new_branch_reason":"No prior New Runtime analytical record was supplied; this begins a distinct record on Fireworks’ serving route and its stated deployment terms for Muse Spark Open."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/muse-spark-open-on-fireworks-meta-s-open-316301/","markdown":"https://newruntime.com/posts/muse-spark-open-on-fireworks-meta-s-open-316301.md","json":"https://newruntime.com/posts/muse-spark-open-on-fireworks-meta-s-open-316301.json"}
related_shifts: ["compute-backed-ai-contracts"]
---

# Muse Spark Open on Fireworks: Meta’s Open Weights Model, Available on US Serverless

## Retrieval answer

**Thesis:** A model’s open weights create options, but the serving route determines which teams can put those options into production.

**Thesis:** A model’s open weights create options, but the serving route determines which teams can put those options into production.

**Mechanism:** A distinct hosting route packages access to open weights with stated geographic capacity, data-retention defaults, and options for production throughput and model adaptation. Fireworks says Muse Spark Open is available through its serverless API on US-only capacity, with Zero Data Retention by default and no US premium. It also presents reserved throughput and serverless LoRA training as routes from initial access toward production use and adaptation. That combination makes deployment terms—not only model weights—a consequential part of the offer.

**Why now:** Fireworks’ announcement adds a provider-specific serving option after earlier reports of the model’s availability on other platforms. For teams evaluating an open-weight model, an additional provider can change the practical choice set, especially when geography, data handling, and customization matter. It does not establish that Fireworks is the best route or that the model will outperform alternatives on a particular workload.

**Evidence boundary:** The evidence is Fireworks’ company announcement. Its performance figures are vendor-reported or benchmark-specific; the source says some third-party results were measured on Meta’s API release. It does not establish comparative performance on a reader’s workload or independently verify service terms. The source reports benchmark results from named evaluators and its own Specialized Intelligence Index, but the results use differing benchmarks and harnesses; the post also says some third-party results were measured on Meta’s API release. Treat performance claims as reported results, not a controlled comparison of Fireworks serving against other providers.

**First implication:** Teams can compare another hosting route against their geographic, data-handling, and workload requirements. A team considering Muse Spark Open should compare the Fireworks endpoint’s current pricing and stated data-handling terms against its own workload requirements, then validate quality and latency on representative tasks.

**First watch condition:** Watch for independent, same-workload tests across providers and confirmation of current pricing and operational terms for reserved throughput and LoRA training. Watch for independently reproducible evaluations of the same workload across providers, alongside confirmation of current pricing and the operational terms of reserved throughput and LoRA training.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Open Models](https://newruntime.com/topics/open-models.md): Explore the open-models topic hub.
- topic hub: [Inference](https://newruntime.com/topics/inference.md): Explore the inference topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)