Skip to content
TrustList
E1
AI Model

Ember-1

New· 4

Fireworks AI's retrained Kimi K3 model that reasons in about 40% fewer tokens, offered as a two-week research preview

About Ember-1

Reasoning models spend a lot of tokens thinking before they answer. Ember-1, published by Fireworks AI on 23 September 2026, is an attempt to cut that cost without changing the answers.

A retrained Kimi K3

Fireworks did not build a new base. It retrained Kimi K3, Moonshot AI's mixture-of-experts model of 2.78 trillion parameters, so that it arrives at the same answers with shorter reasoning. In the developer's own tests that meant roughly 40 percent fewer reasoning tokens. OpenRouter and Vercel's AI Gateway listed it independently within hours of the announcement.

What it accepts and returns

Inputs can be text or images, and it replies in text, with function calling and structured output available. All three listings give the context length as about a million tokens, though they differ slightly, with 1,040K on Fireworks' page and 1,048,576 on the other two.

Results and pricing logic

Against Kimi K3, Fireworks' own table is mixed: a small lead on DeepSWE 1.1 and Terminal Bench 2.1, and a small deficit on SWE-bench Verified. Those figures are Fireworks' own, and when this was read on 25 September 2026 nobody independent had published an evaluation. Per-token rates are the same as Fireworks' rates for Kimi K3, which means the benefit appears as fewer output tokens billed, not a cheaper rate.

Caveats for buyers

Two things matter. Fireworks labels Ember-1 a research preview with serverless access for two weeks, so it could be pulled or repriced. The weights are also not published, which rules out self-hosting.

TrustList summary

Ember-1 is a language model from Fireworks AI, published on 23 September 2026. Its base is Kimi K3, the 2.78-trillion-parameter mixture-of-experts model that Moonshot AI built, and Fireworks retrained it rather than starting from a new base. The goal is to reach similar answers while writing less reasoning.

Features

  • Fewer reasoning tokens: about 40 per cent fewer reasoning tokens than Kimi K3 on Fireworks' own tests, at similar quality.
  • Benchmark results: slightly ahead of Kimi K3 on Terminal Bench 2.1 and DeepSWE 1.1, and slightly behind on SWE-bench Verified, in Fireworks' own table.
  • Inputs and outputs: takes text and images and returns text.
  • Tools and structure: supports function calling and structured output.
  • Research preview: two-week serverless access, and the model may be withdrawn or repriced.

Key facts

  • Developer: Fireworks AI.
  • Context length: about one million tokens. Fireworks lists 1,040K, while OpenRouter and Vercel list 1,048,576.
  • Weights: not published.
  • Listed by: OpenRouter and Vercel's AI Gateway, within hours of launch.

Summary written by TrustList from the business's own published information, Nov 2026.

Benchmarks & AI stats

BenchmarkOfficialCommunity avg
No benchmark scores yet. Be the first to add one.

“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).

Sign in to add a benchmark score for this model.

Ember-1 launches

Trust Score

4/ 100
Trust Score: New

An earned signal from verification, reviews, awards, transparency and engagement - the vendor can't buy it.

Verification
0/100 · 20%
Reviews
0/100 · 30%
Awards
0/100 · 15%
Transparency
18/100 · 20%
Engagement
0/100 · 15%
Joining soon
Recommendations
Coming soon
Complaints
Coming soon

Updated 9/25/2026

Request a demo or quote from Ember-1

Protected by reCAPTCHA - Google Privacy Policy and Terms apply.

By sending, you agree we may share your request and contact details with the provider once you confirm your email.

Reviews

Write the first review of Ember-1

Used it? Your experience helps other buyers decide.

Write a review

Questions & answers

No questions yet. Be the first to ask about Ember-1.

Ember-1 alternatives

Similar listings buyers compare against this one.

Microsoft AI's text-to-speech model for keeping one voice across 23 languages, with voice cloning from short audio

Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions

Microsoft AI's real-time speech-to-text model for voice agents, transcribing 60 languages with continuous language detection

Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0

Liquid AI's open-weight decision model, giving yes or no, choice or score answers from text and images

Cantina Security's first open-weight security model, a post-trained GLM-5.3-Flash for vulnerability research and exploit checks

Black Forest Labs' image generation and editing model with bounding-box layouts, multiple reference images and 4K output

Aleph Alpha's open-weight English-German mixture-of-experts model for self-hosting, released under Apache 2.0 in October 2026