Muse Spark Open: An 816-Billion-Parameter Mixture-of-Experts Model With 42 Billion Active

Muse Spark Open has 816 billion total parameters in a mixture-of-experts design that activates 42 billion per token, with a context window up to 1 million tokens, VentureBeat reported. It accepts text, images, video and audio as input and outputs text only.

In Meta's own table it scores 61.8 on JobBench against GLM 5.3's 61.4 and 89.2 on MCP-Atlas, but trails on Terminal-Bench 2.1 at 76.4 against 83.9 for GLM 5.3 and 85.0 for Kimi K3. Meta reports a 39 on the Artificial Analysis Intelligence Index, against 45 for GLM 5.3 and 40 for Qwen3.8, but VentureBeat could not verify an independent score. Meta calls it the most capable U.S. open-weight model to date.

Source: VentureBeat