---
type: "post"
id: "muse-spark-open-an-open-weight-model-for-ddd53f"
slug: "muse-spark-open-an-open-weight-model-for-ddd53f"
title: "Muse Spark Open, an open-weight model for coding agents"
description: "Thesis: Muse Spark Open treats long-running coding-agent work as a model architecture and serving problem, not merely a matter of extending context length."
retrieval_nugget: "Thesis: Muse Spark Open treats long-running coding-agent work as a model architecture and serving problem, not merely a matter of extending context length."
published_at: "2026-10-15"
updated_at: "2026-10-15"
record_date: "2026-10-15"
date_kind: "published_at"
topics: ["coding-agents"]
source_urls: ["https://www.baseten.co/blog/muse-spark-open-launch"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:muse-spark-open-an-open-weight-model-for-ddd53f"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:ef833452d7179e85559fea635108b703cc6d349d410a02583ea0e8702cfc5d4c","reviewed_at":"2026-10-15T23:07:15.668Z","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":1,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Muse Spark Open treats long-running coding-agent work as a model architecture and serving problem, not merely a matter of extending context length.","observed_facts":[{"text":"Baseten says Muse Spark Open has 816B total parameters, 42.7B active at inference, a 1M-token context window, and accepts text, image, video, and audio input.","source_urls":["https://www.baseten.co/blog/muse-spark-open-launch"]},{"text":"Baseten describes 46 local-attention layers and 16 global-attention layers, and estimates KV-cache use at 128K context falls from 31 GiB to 8.4 GiB per sequence.","source_urls":["https://www.baseten.co/blog/muse-spark-open-launch"]},{"text":"Baseten reports its NVFP4 checkpoint is about 540 GB versus roughly 1.6 TB and fits on one 4×GB300 node.","source_urls":["https://www.baseten.co/blog/muse-spark-open-launch"]}],"mechanism":"Baseten describes a compressed MoE block intended to reduce parameter footprint and inter-GPU traffic, alongside 46 local-attention layers and 16 global-attention layers. At 128K context, the post estimates KV-cache use falls from 31 GiB to 8.4 GiB per sequence.","why_now":"The release makes the weights available and links model design to the repeated inference and memory demands of agent loops.","implications":["The first implication is that teams evaluating open-weight models for coding agents can assess not only task quality but also whether the model’s context and serving footprint fit their workloads."],"evidence_boundary":"The source is Baseten’s announcement; benchmark comparisons are attributed to Meta and use lab-specific agents rather than one shared harness, so they do not establish a controlled ranking across models.","watch_conditions":["What to watch: whether independent teams can reproduce the stated throughput and memory advantages on comparable agent workloads."],"related_records":[],"new_branch_reason":"This begins a distinct analytical record because no prior New Runtime record was supplied for this specific open-weight release and its reported serving architecture."}
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-an-open-weight-model-for-ddd53f/","markdown":"https://newruntime.com/posts/muse-spark-open-an-open-weight-model-for-ddd53f.md","json":"https://newruntime.com/posts/muse-spark-open-an-open-weight-model-for-ddd53f.json"}
related_patterns: ["harness-architecture-outlives-model-choice"]
---

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

## Retrieval answer

Thesis: Muse Spark Open treats long-running coding-agent work as a model architecture and serving problem, not merely a matter of extending context length.

Thesis: Muse Spark Open treats long-running coding-agent work as a model architecture and serving problem, not merely a matter of extending context length. Baseten reports an 816B-parameter model with 42.7B active at inference, a 1M-token context window, and multimodal input; it says the model was co-trained with Muse Code for coding-agent workflows.

Mechanism: Baseten describes a compressed MoE block intended to reduce parameter footprint and inter-GPU traffic, alongside 46 local-attention layers and 16 global-attention layers. At 128K context, the post estimates KV-cache use falls from 31 GiB to 8.4 GiB per sequence. The company also reports that its NVFP4 checkpoint reduces weight storage from roughly 1.6 TB to 540 GB and fits on one 4×GB300 node. These are the reported mechanisms and serving figures, not independently verified measurements.

Why now: The release makes the weights available and links model design to the repeated inference and memory demands of agent loops. The first implication is that teams evaluating open-weight models for coding agents can assess not only task quality but also whether the model’s context and serving footprint fit their workloads. What to watch: whether independent teams can reproduce the stated throughput and memory advantages on comparable agent workloads. Evidence boundary: The source is Baseten’s announcement; benchmark comparisons are attributed to Meta and use lab-specific agents rather than one shared harness, so they do not establish a controlled ranking across models. This begins a distinct analytical record because no prior New Runtime record was supplied for this specific open-weight release and its reported serving architecture.

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: [Coding Agents](https://newruntime.com/topics/coding-agents.md): Explore the coding-agents 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)