Muse Spark
Meta's first proprietary frontier model — natively multimodal reasoning that powers Meta AI across its apps.
About
Muse Spark, released April 8, 2026 by Meta Superintelligence Labs under Chief AI Officer Alexandr Wang, is Meta's first proprietary large language model and the successor to the Llama family. The launch marked Meta's strategic pivot from open weights to closed, product-integrated models after the poor reception of Llama 4.
Spark is a natively multimodal reasoning model accepting text, voice, and image inputs, with visual chain-of-thought, built-in tool use, and multi-agent orchestration. It ships with three reasoning modes: Instant for fast responses, Thinking for reasoning-heavy queries, and Contemplating, a multi-agent mode aimed at long scientific work against Gemini Deep Think and GPT Pro. It powers the Meta AI assistant across meta.ai, the Meta AI app, WhatsApp, Instagram, Facebook, and Messenger.
On the Artificial Analysis Intelligence Index v4.0 it scores 52, fourth among frontier models, while leading all frontier models on HealthBench Hard (medical AI) at 42.8. Coding is an acknowledged gap. The API sits in private preview for select partners; Meta has published no public token pricing.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent (Artificial Analysis Intelligence Index v4.0): 52, behind Gemini 3.1 Pro and GPT-5.4 (57) and Claude Opus 4.6 (53). Independent (HealthBench Hard): 42.8, leading all frontier models vs GPT-5.4 at 40.1.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Release facts (April 8, 2026; MSL; Alexandr Wang; proprietary; three reasoning modes; powers Meta AI; AA index 52; HealthBench Hard 42.8 lead) corroborated across TechCrunch, Fortune, and multiple independent analyses. Docked for: (1) no public API or pricing, so developer evaluation is limited to private preview partners; (2) no public weights and no commitment on ever releasing them; (3) parameter count and context window undisclosed.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 10.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Adoption 8: powers Meta AI across meta.ai, app, WhatsApp, Instagram, Facebook, Messenger; Llama successor
- · Maintenance 10: released 2026-04-08; active flagship with 1.1 release in July 2026
- · Openness 2: proprietary, no public API pricing, weights not released
- · Manual 8: release facts (Apr 8 2026; MSL/Wang; 3 reasoning modes; AA index v4.0 52; HealthBench Hard 42.8 lead) corroborated across TechCrunch, Fortune, independent analyses; docked for private-preview-only API, undisclosed context window and params, no weights