Fireworks launches Ember-1, matching Kimi K3 quality with 40% fewer tokens

Fireworks Research 发布 Ember-1，以更少推理 token 保持 Kimi K3 质量

Fireworks Research released Ember-1, a model built on Kimi K3 that cuts reasoning tokens by 35–50% while keeping accuracy. Across 7 benchmarks and live A/B tests with two customers, quality held. The team ran 50+ training experiments and found K3 spends over 90% of tokens on internal reasoning, much of it unnecessary. Ember-1 preserves useful self-correction and skips unproductive loops. Savings compound in multi-turn agent tasks where prior reasoning is re-read each turn. The model is live on Fireworks' platform as the first in their own model series.

Why it matters: Fireworks distilled Kimi K3 into Ember-1, cutting reasoning tokens by 35-50% with no accuracy drop, backed by 50+ training runs and live customer A/B tests. Score stays below 85 because this is an optimization of an existing model rather than a new capability release, and Fire...