Voices
Muse Spark Benchmarks: GPQA 89.5% Trails Opus 4.6 (92.7%), GPT-5.4 (92.8%), Gemini 3.1 Pro (94.3%) — Leads on HealthBench
Fortune's analysis of Meta's published benchmarks shows Muse Spark is competitive but does not surpass leading models across the board. On GPQA Diamond (PhD-level reasoning), it scores 89.5% — trailing Gemini 3.1 Pro (94.3%), GPT-5.4 (92.8%), and Claude Opus 4.6 (92.7%). However, it outperforms all rivals on HealthBench Hard at 42.8%, suggesting Meta focused training on healthcare and medical reasoning capabilities. Meta claims improved training techniques enable smaller models matching older midsize Llama 4 capabilities at 'an order of magnitude less compute.'
Source
↳ Follow the thread