Muse Spark Open
Muse Spark Open is Meta's first open-weight model from its flagship Muse Spark line. It has 816 billion total parameters with 42 billion active per token and a one-million-token context window. Meta built it for coding, multi-step reasoning, and agentic workflows that call tools, recover from errors, and sustain long-running tasks. At release it was the highest-scoring open-weight model from a U.S. lab on the Artificial Analysis Intelligence Index.
meta/muse-spark-open
Context window
1M
1,048,576 tokens across input and output
Intelligence
39
Artificial Analysis Intelligence Index v4.3
Input
Text
Text output on Doubleword
Released
Oct 2026
Meta · Generation
Architecture: 816B total parameters, 42B active per token.
Reasoning
This model reasons before it responds. Set the depth per request with reasoning_effort: low, medium, high or xhigh (the default). Use high or xhigh for complex coding and agentic tasks, and low for latency-sensitive work. Reasoning can't be fully disabled: none and minimal use the lowest setting (low), and max uses xhigh.
Input: Text only on Doubleword.
API pricing
USD per 1M tokens.
| Tier | Input / 1M | Output / 1M |
|---|---|---|
| Batch (24 hours) | $1.25 | $4.25 |
| Async | $1.25 | $4.25 |
| Realtime | $1.25 | $4.25 |
Run with the OpenAI SDK
Async inference through the OpenAI-compatible API.
from openai import OpenAI
client = OpenAI(
api_key="your-api-key-here",
base_url="https://api.doubleword.ai/v1",
)
response = client.chat.completions.create(
model="meta/muse-spark-open",
messages=[{"role": "user", "content": "Analyze this repository and propose a refactor plan."}],
reasoning_effort="xhigh",
extra_body={"service_tier": "flex"},
)
print(response.choices[0].message.content)