Skip to content
View original post on X: ZyphraOfficial
Published
42/100

Zyphra's ZOE runs 3.1x faster per user than stock SGLang on AgentX

AIZyphra says its ZOE generates 3.1x faster per user than stock SGLang on one 8xMI300X node in the SemiAnalysis AgentX replay of multi-turn coding agent sessions. At 64 concurrent sessions, ZOE reaches 52.8 tokens/s per user at p90 generation speed, versus 16.9 tokens/s for stock SGLang.
ZyphraVerified on XOfficial
@ZyphraAI
View on X

Part of a thread · earlier post

ZOE generates 3.1x faster per user than stock SGLang on one 8xMI300X node using @SemiAnalysis_ AgentX (a replay of multi-turn coding agent sessions).

At 64 concurrent sessions p90 generation speed:

ZOE = 52.8 tokens/s per user
Stock SGLang = 16.9 tokens/s per user

Source details

Zyphra · x.comPublished · added here

Original title: ZOE generates 3.1x faster per user than stock SGLang on one 8xMI300X node using @SemiAnalysis_ AgentX (a replay of multi-turn coding agen...

Topics

AI codingDeployment & engineeringEvals & benchmarks