Fireworks Ember-1 cuts tokens 35% in live tests

Fireworks released Ember-1, a Kimi K3-based model that uses roughly 40% fewer tokens while preserving comparable quality; its Serverless listing prices input, cached input, and output at $3, $0.30, and $15 per million tokens. In two customer production A/B tests, Fireworks reports about 35% fewer tokens per task at comparable quality, though those results are vendor-reported. Teams running long-reasoning coding or tool loops should benchmark Ember-1 before accepting K3’s full bill; the archive’s Fireworks training analysis shows specialization pays only when workload economics are measurable.