Meta Just Made 'Build Your Own Hotel AI' Cheaper to Try

Meta Superintelligence Labs opened its first paid developer API on July 9, pricing Muse Spark 1.1 at $1.25 per million input tokens and $4.25 per million output — undercutting every comparable frontier model that shipped the same month. The API is dual-compatible with both the OpenAI SDK and the Anthropic Messages format, so an existing integration points at Meta by swapping a base URL and key rather than rewriting code, and it ships built-in web-search grounding, returning cited, real-time answers without a developer building a retrieval stack.

That’s a direct input into a debate already live in hospitality tech: Infor’s David Poprawka argued at HITEC 2026 that hotels should stop waiting on vendor roadmaps and build their own agents on their own data, using Infor Portico as a connectivity layer above the PMS, POS, and RMS — and demoed a predictive-housekeeping agent and a $35 million VIP-identification opportunity at a tribal casino as proof it’s more than a slide. The single biggest cost of that pitch has always been the model calls underneath it; a cheaper, drop-in-compatible frontier API from a major lab lowers exactly that barrier, for any operator or vendor tempted to build rather than buy.

It doesn’t resolve the harder problem Terence Ronson’s whitepaper flags: the AI industry has entered what Ronson calls an “engineering-reality phase,” where infrastructure costs — chips, power, cybercrime exposure — show up in vendor pricing and system reliability months before they surface in a PMS release note. A cheaper API makes assembling your own agent fleet more affordable today; it says nothing about whether the underlying compute stays cheap, or whether a property without a data team can operate what it builds. For hoteliers weighing Poprawka’s pitch, Muse Spark’s price tag is a reason to take “build” more seriously — not a reason to stop asking what it costs to run.