Meta closes in on the top with Muse Spark 1.3, and undercuts rivals on price
decoder ·
Meta has released Muse Spark 1.3, its fourth model in five months, with the xhigh tier generally available and a more powerful max version in limited preview. Independent testing shows solid gains on agentic tasks, but the model still trails top performers like Claude Fable 5.1 on most benchmarks. Its main selling point is price: at $0.55 per index task, it is the cheapest model in its performance class. Gains are concentrated in agentic benchmarks, with the model leading on τ³-Banking at 52 percent for the max tier, though it trails on Terminal-Bench and GDPval-AA. Outside agentic tests, it remains mid-pack, improving on GPQA Diamond and CritPt but dropping slightly on AA-LCR and AA-Omniscience factual accuracy. Pricing for the max variant has not been announced, and Meta says larger models and an open-weights release are coming.
decoder