Ember-1 from Fireworks now available on AI Gateway
Ember-1 cuts reasoning tokens by 40% vs. Kimi K3—shorter traces, lower agent costs. Now on Vercel's AI Gateway for coding workflows.
Why it matters
Fireworks' new reasoning model trades marginally on Kimi K3's capability but delivers measurable token savings for agentic applications. For teams running repeated agent calls, the cost reduction and context-window efficiency matter operationally; the two-week preview window and research-only status limit immediate adoption.
The key facts
13 to knowEmber-1 generates ~40% fewer tokens than Kimi K3 at comparable quality (Fireworks evaluation)
1M-token context window, text and image input, tool calling support
Supports implicit prompt caching, Zero Data Retention, No Prompt Training
Available via Vercel AI Gateway with usage tracking, cost budgets, routing rules
Research preview with two-week initial window
Model identifier: fireworks/ember-1
Integrated into Vercel CLI for coding agents
Ember-1 (research preview) reports ~40% fewer tokens than Kimi K3 at comparable quality
1M-token context window; text and image input; tool calling support
Features: implicit prompt caching, Zero Data Retention, No Prompt Training
Available via Vercel AI Gateway with unified usage/cost tracking and routing rules
Two-week research preview window
Integrated into Vercel CLI for coding agent setup
The story so far
Earlier coverage of this storyline
- Introducing Kimi K3 on Amazon BedrockAWS Machine Learning Blog
- MiMo V2.6 models now available on AI GatewayVercel Blog
- This story
Go to the source
Vercel Blogvercel.com
Publisher excerpt: from Fireworks is now available on .Ember-1AI Gateway Ember-1 is a research preview reasoning model built on for coding and agentic workflows.Kimi K3 Fireworks reports approximately 40% fewer generated tokens than Kimi K3 at comparable quality across its evaluations. For coding agents that make…