AI Potluck
Model components / Fine-tuned / chat models

Muse Spark

Meta Superintelligence Labs

Meta Superintelligence Labs' flagship closed-weight API model, launched April 2026. Closed top-5 on the May 2026 LMArena text leaderboard (Elo 1489), the first non-Anthropic/Google/OpenAI model to break the top 5 in 2026. Represents Meta's parallel closed-weights track alongside the open-weights Llama line (Llama 5, April 2026), effectively splitting Meta into two flagship brands. Chat/instruction-tuned tier sits in Cat 3 alongside other API-only frontier models.

Muse Spark, first model of Meta Superintelligence Labs' new 'Muse' series; released 8 Apr 2026. Natively multimodal reasoning model with tool-use, visual chain-of-thought, and multi-agent 'Contemplating' mode. Proprietary/API-only: powers the Meta AI assistant (meta.ai + Meta AI app), private API preview to select users; no open weights. Verified via Meta's own blog June 2026.

Openness

1 high confidence
1.0
weights
closed
data
closed
code
closed
license
Proprietary(meta.ai/Meta AI app + private API preview

No weights released, a closed assistant-surface model; only a private API preview. Fully closed.

Adoption

4 low confidence
4.0

Powers the Meta AI assistant on meta.ai and the Meta AI app (rolling out to WhatsApp/Instagram/Facebook/Messenger/AI glasses), an enormous consumer surface, but Meta discloses no Muse Spark-specific user count, and at scoring time it is not yet the model behind the full family of surfaces. Placed at 4 on the consumer-app surface it powers; not 5 because no standalone figure and rollout incomplete. Attributed honestly to the surface, not the model.

Capability

5 medium confidence
5.0

Bumped 4 -> 5: the independent confirmation the prior note awaited now exists - LMArena ranks Muse Spark #6 overall (top of the non-Anthropic/Google frontier), corroborating Meta's vendor-reported reasoning scores.

Unchanged since 2026-06-24 (last edited, not re-checked)