Artificial Analysis lists Gemma 4 E2B
Artificial Analysis has published a page for Gemma 4 E2B, an Apache 2.0 open-weights model that activates 2.3B of its 5.1B parameters per token and accepts text, image, speech and video input.
Benchmarking firm Artificial Analysis has published a release page for Gemma 4 E2B, the newest small model in the Gemma family. The page sets out the model's specs and will track its intelligence, cost and speed.
According to the listing, Gemma 4 E2B has 5.1 billion total parameters, of which 2.3 billion are active per token during inference. It has a 128k-token context window, which Artificial Analysis puts at about 192 A4 pages. It accepts text, image, speech and video as input, and the page lists text and image as output modalities. The weights are on Hugging Face under the Apache 2.0 license.
The release covers two models, and Artificial Analysis says each has its own intelligence, performance and pricing profile. The page does not name the two variants or say how they differ.
Artificial Analysis compares open-weights models only with others in the same size class. Its classes are "tiny" (up to 4B parameters) and "small" (4B–40B parameters). The page does not say which class Gemma 4 E2B falls into. The firm compares reasoning models with both reasoning and non-reasoning models, and non-reasoning models only with each other.
The page says its capability indexes draw on evaluations including AA-Omniscience v1.1, AA-Briefcase v1.1, HLE-Diamond, AutomationBench-AA, GDP.pdf, Terminal-Bench 4.0 and FrontierCode-AA 2.0. The material reviewed has no actual scores, prices, speed or latency figures for the model, so it is not yet possible to say how Gemma 4 E2B compares with its peers.