Meta Muse Spark 1.3: Half the Price of Rivals, Almost at the Top

Meta has released Muse Spark 1.3 — already the fourth version of its flagship language model in five months. The model scores 61–62 on the overall Intelligence Index (up from 53 in July), hits 94% on the GPQA Diamond benchmark for complex scientific tasks, and 85–86% on Terminal-Bench.

The key differentiator is price: $0.55 per standard index request versus $0.94–$1.23 for the closest competitors at the same performance level. Input token cost is $1.25 per million; output is $4.25 per million. The max version is available in limited preview and uses 62% more reasoning tokens.

In five months, Meta has gone from an also-ran to second place: Claude Fable 5.1 still leads on Terminal-Bench (91% vs. 86%), but the gap narrows with every release. Meta also announced open weights for the model — already in development.