Fireworks Forge adds Ultra-fast tier and half-price Background Mode

Fireworks AI announced a set of inference and training products at Forge, its first annual conference, on November 3, 2026. The launches include a faster serving tier, a half-price mode for work that can wait, and serverless hosting for fine-tuned GLM 5.3 Flash adapters.

What changed

• Ultra-fast is in public preview, running on NVIDIA Blackwell GPUs. Fireworks says it reached up to 1,500 tokens per second on DeepSeek V4.1 Flash, and that a 30-turn agent task in its benchmark fell from 1 minute to 8 seconds.
• Background Mode runs eligible work that can wait, such as eval data generation and overnight research, at half price. Fireworks also added multi-day prompt caching for repeated context.
• Higher serverless rate limits and Elastic Burst are generally available. Elastic Burst lets customers go past their usual ceiling when spare capacity exists, with no code or pricing changes.
• Specialized Model Inference is in public preview: serverless, per-token serving of LoRA adapters, starting with GLM 5.3 Flash, with no serving charge while a model sits idle.
• Fireworks Sovereign, in private preview, runs Fireworks' inference engine in a customer's own cloud, neocloud or data center.

What it means

Background Mode and Elastic Burst change the cost and capacity math for batch and bursty agent traffic on Fireworks. Teams that fine-tune GLM 5.3 Flash can now serve adapters per token instead of sizing a dedicated deployment. Fireworks did not publish Ultra-fast pricing in the announcement.

The same base models on RouterPlus

Fireworks is not a provider in the RouterPlus catalog. Lowest prices for the two models Fireworks named, RouterPlus catalog, November 4, 2026, per million tokens:

If you need reserved capacity for rollouts or batch work, tell us what you need.

Sources

1. Fireworks AI: Fireworks unveils new products at Forge (November 3, 2026)
2. RouterPlus model catalog