Ember-1 from Fireworks now available on AI Gateway
Fireworks’ Ember-1 reasoning model is now accessible via Vercel’s AI Gateway, offering shorter reasoning traces and a 1M-token context window for coding agents and workflows.
Ember-1, a research-preview reasoning model from Fireworks built on Kimi K3, is now available on Vercel’s AI Gateway. The model targets coding and agentic workflows, delivering approximately 40% fewer generated tokens than Kimi K3 at comparable quality in Fireworks’ evaluations. Shorter reasoning traces can lower output costs and reduce context carried forward in multi-step agent calls, improving efficiency for repeated model interactions.
Ember-1 supports a 1 million-token context window and accepts both text and image inputs. It includes tool calling capabilities and implicit prompt caching, enabling agents to manage longer sequences without manual prompt management. The Fireworks endpoint enforces Zero Data Retention and prohibits prompt training, aligning with privacy-focused deployment needs.
The model is available as a research preview with an initial two-week window. Users can provision or reuse an AI Gateway API key, and the system automatically detects installed agents and configures their connection to the gateway. Selecting fireworks/ember-1 in the agent’s model configuration activates the model within the gateway.
AI Gateway provides a unified API for model calls, featuring built-in usage and cost tracking, budget controls for API keys, and configurable routing rules. Ember-1’s integration streamlines deployment for developers using Vercel’s platform while maintaining cost and privacy controls.