fal and Fireworks AI start early launch of EAGLE-4 inference accelerator
fal said it worked with Fireworks AI on the early launch of EAGLE-4, describing it as its fastest inference version yet after the EAGLE-1 to EAGLE-3 line. The company reported 961.6 tokens/s on Alibaba Qwen's Qwen3.8-27B using TokenSpeed in BF16 on four NVIDIA B300 GPUs with TP4, a single request and temperature 0. fal pointed readers to a technical blog on how EAGLE-4 works for creative workflows and to a GitHub repository; the performance figures are fal's own and have not been independently verified.
fal presents EAGLE-4 as its fastest inference release so far and ties the reported throughput on a Qwen model to creative workflows, giving practitioners a vendor-stated data point on serving speed.
We are excited to work together with @FireworksAI_HQ on the early-launch of EAGLE-4. Check out our technical blog to understand how EAGLE-4 works and why it matters for creative workflows: https://blog.fal.ai/eagle-4-technical-deepdive/
Eagle: EAGLE-4 is here. From EAGLE-1 to EAGLE-3, we’ve kept making inference faster. EAGLE-4 is our fastest yet. 961.6 tokens/s on @Alibaba_Qwen’s Qwen3.8-27B with TokenSpeed (BF16, 4× @NVIDIA B300, TP4, one request, T=0). Github: https://github.com/eagle-hq/EAGLE-4
#fal#Fireworks AI#EAGLE-4#Qwen3.8-27B#Alibaba Qwen#NVIDIA B300#TokenSpeed