Skip to content
View original post on X: GoogleOfficial· 46/100AI score46/100

Gemini 4 Argon speeds up Gemma 4 E2B inference 2.85x on A100

AISummary

Google says its Gemini 4 Argon optimization raised Gemma 4 E2B inference on an A100 via vLLM from about 37,900 to over 108,000 tokens per second. The post reports a 2.85x speedup with 100% token parity and no quality drop, demonstrated by generating 64 live Python simulations at once.

Post on XView on X
GoogleVerified on X
@Google

Part of a thread · earlier post

Gemini 4 Argon is also speeding up our @GoogleGemma open models — making them nearly three times as fast.

In just a few days, Argon optimized Gemma 4 E2B to achieve a 2.85x inference speedup on an A100 via vLLM, jumping from ~37,900 to over 108,000 tokens per second with 100% token parity and zero quality drop.

Here’s that speedup in action, generating 64 live Python simulations at once ↓

https://x.com/osanseviero/status/2110096066293207535/video/1

Source: Google · x.comPublished · added here

Topics

Google / GeminiOpen-source ecosystemDeployment & engineeringProduct updatesEvals & benchmarks