Ember-1 from Fireworks now available on AI Gateway
Source
vercel.com
Author
Zachary Chen
Date
Context
Vercel added Fireworks' Ember-1 research preview to AI Gateway, its shared interface for calling models and tracking usage. Developers could select fireworks/ember-1 through that interface. The announcement also describes a setup command that detects installed coding agents, creates or reuses gateway access credentials, and configures their connection.
This gives teams a way to evaluate the model through existing gateway budgets and cost tracking. The launch specified an initial two-week preview window, so it establishes a trial offering rather than permanent availability. The model announcement explains the training rationale.
Key takeaways · AI-distilled
Fireworks built Ember-1 on top of Kimi K3 as a reasoning modelA model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.Full definition → for coding and agent workflows; the roughly 40% token reduction versus Kimi K3 is Fireworks' own reported figure from its evaluations, not an independent benchmark.
Ember-1 supports a 1M-token context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition →, text and image input, tool useA model's ability to call external functions — run code, search the web, edit files — instead of only generating text.Full definition → and implicit prompt cachingReusing the model's processed form of a repeated prompt prefix so subsequent calls skip re-reading it, cutting cost and latency substantially.Full definition →, and the Fireworks endpoint offers Zero Data Retention and no training on prompts.
It ships as a research preview with an initial two-week window; developers call it as fireworks/ember-1, and running vercel ai-gateway setup detects installed coding agents and configures them to route through AI Gateway.
Terms in this piece · Glossary
reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
prompt caching — Reusing the model's processed form of a repeated prompt prefix so subsequent calls skip re-reading it, cutting cost and latency substantially.
