Particle.news

Meta Unveils Muse Spark 1.1 as Low‑Cost Agentic and Coding Model

The company is offering a public API preview with aggressive per‑token pricing to push enterprise adoption and pressure rivals.

Overview

  • Meta launched Muse Spark 1.1 through a public Meta Model API in a U.S. preview, and new accounts receive $20 in trial credits before pay‑as‑you‑go billing.
  • The company set pricing at $1.25 per million input tokens and $4.25 per million output tokens to undercut higher rates from OpenAI and Anthropic.
  • Meta says Spark 1.1 is built for agentic tasks and coding with a 1 million‑token context window, the ability to run parallel sub‑agents, and training to use desktop, mobile, and browser tools.
  • Meta published internal benchmarks claiming Spark 1.1 outperformed certain Opus and GPT‑5.5 variants on agentic tests but those results are company‑reported and lack independent verification.
  • Investors reacted positively with a stock rally this week, and Meta says it will eventually offer an open‑source variant while pursuing in‑house chips and cloud capacity to lower long‑term token costs.