Meta Superintelligence Labs will reach genuine frontier parity (within about 2 points on Artificial Analysis's Intelligence Index of the top closed model) within its first two model generations, with its reported agentic-benchmark gains holding up under independent, non-Meta-sourced testing.

confidence 0.4 · horizon 2027-06-30 · status open

Falsifiers

Independent, non-Meta testing of Muse Wonder Full's Terminal-Bench or other agentic scores shows a material drop (10+ points) from Meta's reported 57%
The next Meta Superintelligence Labs flagship shows no further narrowing of the Intelligence Index gap to the closed frontier leader

-1 Meta releases Muse Spark Open, an 816B-parameter open-weight model built to run on Mac — Described as the first base model trained under Meta Superintelligence Labs, Muse Spark Open scores 39 on Artificial Analysis's Intelligence Index — far below the ~58 closed frontier and weak on agentic evaluations specifically — though this is Meta's open Spark line rather than the higher-scoring closed Wonder line the thesis was adopted on, so it only weakly speaks to the flagship trajectory question.