Understand this story
On November 3, 2026, Fireworks released multiple products at its inaugural Forge conference, covering inference acceleration, model routing, training, and enterprise deployment. Fireworks Ultra-fast, based on NVIDIA Blackwell GPUs, entered public preview, demonstrating a speed of 1,500 tokens/s on DeepSeek V4.1 Flash and reducing the completion time of a 30-turn agent task from 1 minute to 8 seconds; Background Mode offers a half-price option for running tasks that can be deferred. Fireworks Nexus added a desktop client and can be used in Codex, Claude Desktop, and OpenCode, with the Pro version starting at $8 per month and the Teams version starting at $10 per seat per month. On the training side, No Eval entered private preview, and Specialized Model Inference serves LoRA adapters in a serverless manner, debuting with GLM 5.3 Flash; Fireworks Sovereign entered private preview and can deploy the inference engine to customers' own infrastructure. The next day, Unite.AI reported that Fireworks Sovereign officially stated that throughput on the same hardware increased by 20% to 50%.
AI-generated from public reports
Latest development · Report
Fireworks Releases Ultra-fast Inference Preview and Sovereign Private DeploymentDevelopments
1 confirmed developments
- Source:Fireworks AI
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 4
Unite.AI
Fireworks Releases Ultra-fast Inference Preview and Sovereign Private Deployment
At its inaugural Forge conference, Fireworks released multiple enterprise AI products, among which the Ultra-fast inference preview is said to reach up to 1500 tokens/s on DeepSeek V4.1 Flash, cutting the completion time of a 30-turn agent benchmark from one minute to eight seconds. For deferrable tasks, Background Mode runs at half price; Nexus supports routing models in Codex, Claude Desktop, and OpenCode, with the Pro version starting at $8 per month. Fireworks Sovereign deploys the inference engine to customers' own infrastructure as a private preview, and the company says throughput on the same hardware improves by 20% to 50%.
Nov 3
Fireworks AIFirst-party
Fireworks unveils multiple products including Ultra-fast, Nexus, and Sovereign at the Forge conference
At its inaugural Forge conference, Fireworks released multiple products spanning inference acceleration, model routing, training, and enterprise deployment. Fireworks Ultra-fast, built on NVIDIA Blackwell GPUs, has entered public preview, demonstrating speeds of 1500 tokens/s on DeepSeek V4.1 Flash, cutting the completion time of a 30-turn agent task from 1 minute to 8 seconds; Background Mode offers a half-price option for running deferrable tasks. Fireworks Nexus adds a desktop client usable in Codex, Claude Desktop, and OpenCode, with the Pro version starting at $8 per month and the Teams version starting at $10 per seat per month. On the training side, No Eval has entered private preview, and Specialized Model Inference serves LoRA adapters in a serverless manner, debuting with GLM 5.3 Flash; Fireworks Sovereign has entered private preview and can deploy the inference engine to customers' own infrastructure.
Heat trend for this story
- Current
- 11
- Period peak
- 14 Nov 4 08:00
- 24-hour change
- —
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.