Muse Spark 1.3

Muse Spark 1.3 is a chat model attributed to Meta in our catalog. It is listed as a conversational text model, meaning it is designed to take a prompt or multi-turn dialogue and return generated text. Our database entry for this model is intentionally narrow: beyond the creator and model type, the details we have positively confirmed are the independent performance measurements described below. Where a specification is not listed here, it means we have not verified it rather than that the model lacks it. The performance data we track for Muse Spark 1.3 comes from Artificial Analysis. In those measurements, the model produced roughly 110.75 output tokens per second, which places it in a reasonable range for interactive streaming output — text appears faster than most people read. The measured time to first token was approximately 39,264 milliseconds, or just under 40 seconds. That is a very long lead time before the first token appears, and it is the single most distinctive number in this model's profile. Long first-token latency of this magnitude typically reflects extended pre-generation processing, queueing on the measured endpoint, or a serving configuration that batches heavily; the underlying cause is not something we can confirm from the data we hold. In practical terms, that latency profile matters more than raw throughput for deciding where Muse Spark 1.3 fits. Once generation begins it streams at a competitive rate, so long-form output is not the bottleneck — waiting for generation to start is. Workloads that run asynchronously, in batches, or behind a job queue absorb that delay without user-visible cost. Workloads where a person is watching a cursor blink generally do not. We do not currently track a confirmed context window, modality set, or benchmark accuracy scores for this model; readers evaluating it for a specific task should check the provider's own documentation alongside the pricing table on this page.