Meta launched Muse Spark on April 8, its first absolutely closed AI mannequin, abandoning its open-source Llama technique
Abstract
- Meta launched Muse Spark on April 8, the primary product from its Meta Superintelligence Labs unit, constructed from scratch by Alexandr Wang’s crew after a $14.3bn Scale AI deal.
- The mannequin is absolutely proprietary with no open weights, a direct reversal of the Llama technique that reached 1.2 billion downloads by early 2026.
- Meta inventory rose 9% on launch day, and the mannequin will roll out throughout WhatsApp, Instagram, Fb, and Messenger within the coming weeks.
Meta launched Muse Spark on April 8, its first absolutely closed AI mannequin, abandoning its open-source Llama technique. The launch marks the inaugural product from Meta Superintelligence Labs, the unit constructed round Alexandr Wang after Meta’s $14.3bn funding in Scale AI.
Wang stated in an announcement: “9 months in the past, we rebuilt our AI stack from scratch. New infrastructure, new structure, new knowledge pipelines. That is the 1st step. Greater fashions are already in improvement with plans to open-source future variations.”
In contrast to Llama, Muse Spark’s weights will not be publicly accessible. API entry is presently by invitation solely, concentrating on choose companions. Meta has stated it hopes to open-source future variations, framing the present closure as short-term.
Gartner analyst Arun Chandrasekaran described the transfer as a “main shift,” saying it indicators Meta’s intention to maneuver away from the Llama model completely.
What Muse Spark truly does
The mannequin is natively multimodal, dealing with textual content, picture, and voice inputs. Its flagship function is a “Considering” mode that runs a number of reasoning brokers in parallel earlier than responding, competing straight with Gemini Deep Suppose and GPT Professional.
Meta collaborated with over 1,000 physicians to curate health-related coaching knowledge, and the mannequin is being marketed as a private well being reasoning software alongside its normal assistant capabilities.
Muse Spark sits beneath GPT-5.4 and Gemini 3.1 Professional on the Synthetic Evaluation Intelligence Index, scoring 52 in opposition to their 57. Meta has not disclosed the mannequin’s parameter depend or structure particulars. As crypto.information reported, the mannequin beat Gemini 3.1 Professional on a number of health-related benchmarks that Meta prioritized in its analysis suite.
Why Meta made the swap now
As crypto.information documented, Meta had been signaling a phased strategy to its subsequent AI technology, holding core parts proprietary whereas assessing security dangers.
The swap to a completely closed first launch displays the aggressive stress from OpenAI and Anthropic, each of which have proprietary fashions producing billions in API income that Meta’s open-source strategy couldn’t seize.
Meta’s 2026 capital expenditure is guided at $115bn to $135bn, almost double 2025 ranges. Meta inventory rose greater than 9% on launch day, the strongest single-day response to a Meta product announcement in over two years. The developer neighborhood that constructed on Llama is now being requested to attend for a future open-source launch with no confirmed timeline.
