Meta’s Muse Beats GPT-6 Astra and Claude on Memory and Coding Tests

Benchmark wins matter less than they appear when independent rankings still put a rival model first — the gap between vendor-selected tests and neutral leaderboards is the real story here. For builders choosing a frontier model, that divergence complicates any simple 'best model' claim. The absence of release date and pricing leaves the commercial picture entirely open, and whether Muse's benchmark edge survives third-party evaluation is the question worth watching.