Today ainotis

Models28 Sep 2026That day's edition

Fireworks says its Ember-1 research preview delivers Kimi K3 quality with 40 percent fewer tokens.

Fireworks reports reductions against Kimi K3 at max effort of 51.9 percent on Terminal Bench 2.1 and 15.5 percent on SWE-bench Verified in its own table, in a two-week Serverless preview; the post gives no Ember-1 price.

Check our sources · 1 source, 3 claims

Your reaction

One press adds one. We count a number for each notice and day, never who pressed it.

Check our sources

1 source, 3 claims. We opened the source and checked every sentence above against it.

1 Introducing Ember-1Fireworks AI · 23 Sep 2026 · 3 claims Open the source

  1. Fireworks published Ember-1 on September 23, 2026, describing it as delivering Kimi K3 quality with 40 percent fewer tokens.

    PUBLISHED 9/23/2026 ... Ember-1 is a new specialized model from Fireworks Research that delivers Kimi K3’s quality with 40% fewer tokens.

  2. In Fireworks own comparison against Kimi K3 at max effort, Ember-1 shows reductions of 51.9 percent on Terminal Bench 2.1, 23.7 percent on DeepSWE 1.1 and 15.5 percent on SWE-bench Verified.

    Narrowed to what the source supports
    Ember-1 vs. K3 Max: Terminal Bench 2.1 -51.9% / -23.1 USD; SWE-bench Verified -15.5% / -68.1 USD; DeepSWE 1.1 -23.7% / -126.9 USD

  3. Ember-1 is rolling out as a Research Preview on Fireworks Serverless alongside Kimi K3, with two-week access that Fireworks says may become permanent based on demand.

    Narrowed to what the source supports
    We computed per-benchmark cost using the public Kimi K3 API pricing (uncached input $3/M tokens, cached input $0.30/M, output $15/M) ... rolling out as a serving option alongside the base Kimi K3 model as a Research Preview release on Serverless ... two-week serverless access to new research models, making them permanent based on community demand

Nothing appears on this site that we have not opened and linked.

Filed under