November 3, 2026 · Fireworks AI
Ultra-fast Serverless mode in Preview, with DeepSeek V4.1 Flash 1017 at 6-8x speed
Ultra-fast is a new high-speed Serverless mode, now in Preview, for highly latency-sensitive applications. Ultra-fast variants aim for 6-8x generation speed, and up to 10x for agentic coding tasks. Like Fast, it is not a different model, so output quality is the same.
Ultra-fast is available for DeepSeek V4.1 Flash 1017. To use it, set model to accounts/fireworks/routers/deepseek-v4p1-flash-1017-ultrafast. Priority tier is not available for Ultra-fast.
Ultra-fast is priced at a premium and has lower rate limits than the standard model. Preview modes are under active development and may change quickly. See Serverless Modes, Serverless pricing, and Serverless rate limits.
Latest Fireworks AI platform changes
Public cookbook moves to fw-ai-external/cookbook, fireworks-training-cookbook on PyPI
DeepSeek V4 Flash (0731) removed from the shared Serverless Training pool
Rune 26B-A4B available on Serverless, returning probabilities for up to 256 questions
fireworks-ai 1.2.30 replaces TITO sidecar 502 errors with source-specific error codes
Six Serverless models deprecated, to be decommissioned on November 7, 2026