Fireworks expands fast and sovereign inference for long-running agents
Fireworks announced an Ultrafast preview claiming more than 1,500 tokens per second, multi-day prompt caching, background mode and managed Sovereign deployment on customer-owned GPUs.
AT A GLANCE
Key points
- Ultrafast claims inference above 1,500 tokens per second.
- Multi-day caching and background mode target long agent jobs.
- Sovereign runs managed inference on customer GPUs.
What is changing?
At its November 3 Forge event, Fireworks expanded its enterprise inference capabilities. The company says its Ultrafast preview on NVIDIA Blackwell exceeds 1,500 tokens per second for interactive applications. This is a provider claim, not an independently verified benchmark.
For long-running agents, multi-day prompt caching is intended to reduce the cost of reusing shared context, while background mode targets work that does not require an immediate response. Fireworks also announced higher rate limits and cold starts under one minute.
The Sovereign option runs Fireworks’ inference engine on GPUs in the customer’s cloud or data center. Customers retain network, access and change policies, while Fireworks manages upgrades, monitoring and incident response.
Explore the original sources
Source date: 4 November 2026.
Published by AI Radar: 4 Nov 2026, 01:14 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…