# Fireworks opens faster inference preview and adds half-price background tasks

_Published Tuesday, November 3, 2026 at 5:19 PM EST · AI · Latest · Tier 2 — Notable_

![Fireworks opens faster inference preview and adds half-price background tasks — Primary](https://www.unite.ai/wp-content/uploads/2026/11/fireworks-announces-faster-agent-inference-and-sovereign-deployment.jpg)

Fireworks opened a public preview of Ultra-fast inference on NVIDIA Blackwell GPUs on November 3, saying it reached up to 1,500 tokens per second on DeepSeek V4.1 Flash. The company said its 30-turn agent benchmark finished in eight seconds instead of one minute.

Background Mode runs eligible delay-tolerant tasks at half price. Fireworks also introduced a public preview that serves customized model adapters on shared infrastructure, charging per token without deployment or idle-time fees; training costs extra.

A separate private preview, Fireworks Sovereign, runs inference and training inside customer-owned environments. Fireworks operates the serving software while customers retain infrastructure policies. Shared adapter serving carries no preview latency guarantees.

## Sources

- [Unite.AI](https://www.unite.ai/fireworks-announces-faster-agent-inference-and-sovereign-deployment/)

---
Canonical: https://techandbusiness.org/newswire/keBneNsI81fhUP4--Czxie
Published: 2026-11-03T22:19:29.915Z
Story chronology: 2026-11-03T21:16:02.000Z
Retrieved: 2026-11-04T00:45:37.529Z
Publisher: Tech & Business (techandbusiness.org)