Mark Zuckerberg’s X post introduced what Meta is calling “Muse Spark 1.1,” a model the company describes as “strong” at agentic and coding tasks and available through its new Meta Model API as well as in Meta AI.
In the post, Zuckerberg claims Muse Spark 1.1 is strongest in “agentic performance, tool use, and computer use,” and adds that the model handles long-running tasks with a 1M token context window, can delegate work to sub-agents running in parallel, and is trained to use computer interfaces on desktop, mobile, or browser. Meta also portrays the API as a first step toward letting developers build with Muse Spark, alongside a broader push for “strong agentic and multimodal models at very low cost.”
The benchmark table attached to the announcement compares Muse Spark 1.1 with Muse Spark, Gemini 3.1 Pro (high), Opus 4.8 (max), and GPT 5.5 (xhigh) across agentic, coding, and multimodal tests. On Meta’s chart, Muse Spark 1.1 posts its strongest numbers on MCP Atlas for “scaled tool use” at 88.1, OSWorld-Verified for “agentic computer use” at 80.8, Terminal-Bench 2.1 for “agentic terminal coding” at 80.0, and CharXiv Reasoning for “chart QA” at 88.4. It also shows 61.5 on SWE-Bench Pro and 76.3 on BabyVision, while some of the competing models still appear ahead on selected benchmarks.
A user reply highlighted pricing as $1.25 for input, $0.15 for cached input, and $4.25 for output, suggesting Meta is positioning the model aggressively on cost. The announcement also drew a stream of jokes and jabs on X, including remarks about Zuckerberg posting from his @finkd account and Elon Musk replying, “Jinx.”
Source: X
