Model Overview & Capabilities
Muse Spark Open is Meta’s flagship open-weights 816B Mixture-of-Experts model (activating 42 billion parameters per token) released on October 15, 2026, holding Rank #14 on our leaderboard with an index score of 42.5. Tested under canonical harnesses, Muse Spark Open scores 43.0% on SWE-bench Verified, 42.0% on Terminal-Bench 2.0, 43.0% on Aider Polyglot, and 83.0% on GPQA Diamond graduate reasoning.
The model is distributed under the Apache 2.0 license for unrestricted self-hosting and fine-tuning, while hosted cloud providers offer managed API access at $1.25 per million input tokens and $4.25 per million output tokens, giving an effective blended rate of $2.00 per million tokens under standard 3:1 input:output weighting. This yields a highly competitive value metric of $0.0471 per index point for engineering teams requiring managed enterprise cloud infrastructure.
Featuring a 1,048,576-token context window, generation throughput of 115 tokens per second, and an initial response latency of 1.10 seconds, Muse Spark Open empowers engineering organizations with full model weight ownership, on-premises deployment security, and flexible domain-specific fine-tuning for proprietary enterprise software codebases and high-security compliance environments.
