AI BriefingFireworks Releases Ember-1 Reasoning Model on AI Gateway
Key point
Ember-1 is a research preview model built on Kimi K3 that reportedly generates 40% fewer tokens than its base model.
Details
Fireworks has released Ember-1, a research preview reasoning model now available on AI Gateway. Built on Kimi K3, Ember-1 is optimized for coding and agentic workflows, with Fireworks reporting approximately 40% fewer generated tokens than Kimi K3 at comparable quality. This reduction in reasoning traces helps lower output costs and context usage for coding agents that make repeated model calls.
The model supports a 1M-token context window, text and image input, tool calling, and implicit prompt caching. The Fireworks endpoint includes Zero Data Retention and No Prompt Training guarantees. Ember-1 is available as a research preview for an initial two-week window.
Developers can access Ember-1 by using the model name fireworks/ember-1 across APIs and coding agents. For coding agents, the Vercel CLI can automatically detect installed agents, provision an AI Gateway API key, and configure the connection. AI Gateway provides a unified API with built-in usage tracking, cost budgets, and routing rules.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Rootless Linux Sandbox 'Drop' Released for Isolating AI Agents and Third-Party Programs
Perplexity: AI Agents Bypass SPACE Network Policies; Direct Tests Reveal Weaknesses in 7 of 9 Third-Party SandboxesFireworks Releases Ember-1: Matches Kimi K3's Performance with 40% Fewer Tokens