Muse Spark Open is an open-weight model from Meta optimized for coding. It is trained for agentic coding workflows, in which a model reads a codebase, calls tools and acts on what they return, and Meta positions it on two properties: reliable tool calling and deep codebase understanding. It was released on 15 October 2026 and is available on AI/ML API as meta/muse-spark-open.
The model belongs to Meta's Muse Spark line. Muse Spark 1.3 is the general-purpose API model of that line, and Muse Wonder is Meta's frontier agentic model; Muse Spark Open is the variant published with open weights and tuned for coding.
The model targets coding agents rather than single-turn code completion. A 1M-token context window lets an agent keep a large part of a repository, its tool outputs and the conversation history in one request, and reasoning tokens let the model plan a change before it makes it. Image, video and PDF input cover the material that sits around code: screenshots of a UI, recordings of a bug, design documents and specifications.
Input is billed at $1.71925 per 1M tokens, cached input at $0.20631 per 1M tokens and output at $5.84545 per 1M tokens. Reasoning tokens are billed at the output rate. This is the same rate as Muse Spark 1.3 on AI/ML API.
Coding agents that work across a repository: reading the relevant files, calling tools such as a shell, a test runner or a search index, and editing code based on what those tools return.
Large-context code tasks, such as reviewing a long diff, answering questions about an unfamiliar codebase or planning a refactor with most of the project in the prompt.
Debugging from mixed inputs, where the model gets a stack trace together with a screenshot, a screen recording or a PDF specification of the expected behaviour.
Teams that want an open-weight Meta model for coding work and use the hosted endpoint for development or overflow capacity.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "meta/muse-spark-open", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "meta/muse-spark-open", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"meta/muse-spark-open","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
Billed per token: $1.71925 / 1M input tokens, $0.20631 / 1M cached input tokens and $5.84545 / 1M output tokens. Reasoning tokens are billed as output.
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Intelligence | 38.9 | Composite score across standardised reasoning, knowledge and problem-solving evaluations, measured independently by Artificial Analysis | Source | October 16, 2026 |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Muse Spark Open This page | Agentic coding and codebase tasks | |||
| Agentic workflows and structured output | ||||
| Long-running agents, computer use and large-repo coding | ||||
| Long-context multimodal reasoning and tool use |
Yes. Muse Spark Open is live on AI/ML API as meta/muse-spark-open, through the chat completions endpoint.
Muse Spark Open has a 1,048,576-token context window and can return up to 943,718 tokens.
It accepts text, images, video, audio and PDF documents and returns text. Reasoning, tool calling and streaming are supported.
Muse Spark Open is the open-weight member of Meta's Muse Spark line and is trained specifically for agentic coding, with an emphasis on tool calling and codebase understanding. Muse Spark 1.3 is the general-purpose model of the line. On AI/ML API both are billed at the same per-token rate.
Muse Spark Open is priced at input $1.71925 / 1M tokens, cached input $0.20631 / 1M tokens and output $5.84545 / 1M tokens. Reasoning tokens are billed at the output rate.
Meta trained it for agentic coding: multi-step work in which the model reads a codebase, calls tools and edits code based on the results. Its 1M-token context window fits a large part of a repository in a single request.
