Atlas

Models

← All models

Gemma 4 E2B IT

Google DeepMind · 2026-04-02 · 5.1B parameters

Gemma 4 E2B IT is Google DeepMind's instruction-tuned, dense, edge-oriented Gemma 4 checkpoint, released on April 2, 2026 under Apache 2.0; the official checkpoint repository reports 5,123,178,051 total parameters. Per-layer embeddings give it an approximately 2.3-billion-parameter effective footprint rather than a sparse active-parameter count; it accepts text, images, audio, and video frames, generates text across a 131,072-token context, and offers caller-controlled thinking. Google states the pretraining-data cutoff only as January 2025, so the structured knowledge cutoff remains null.

Benchmark scores

BenchmarkScore
AA-LCR15.0
AA-Omniscience Index-24.0
Artificial Analysis Agentic Index1.5
Artificial Analysis Coding Index7.2
Artificial Analysis Intelligence Index9.7
Artificial Analysis Omniscience Accuracy8.0
Artificial Analysis Omniscience Hallucination Rate32.9
Artificial Analysis Openness Index38.9
Capability123.5
CritPt0.0
GDPval-AA v287.4
GDPval-AA v2 (normalized)0.0
GPQA Diamond43.3
Humanity's Last Exam (Text-Only)4.8
IFBench (Artificial Analysis)38.0
MMMU Pro44.6
SciCode (Artificial Analysis)20.9
Terminal-Bench 2.10.4
Terminal-Bench Hard3.0
τ²-bench (Telecom)22.2
𝜏³-Banking4.5
Loading Atlas data…