Google DeepMind's edge-optimized model in the Gemma 4 family. With 4.5B effective parameters (using Per-Layer Embeddings), it natively processes text, images, and audio. 128K context, 140+ language support, Apache 2.0 license. Released April 2, 2026.
Parameters
4.5B effective (8B total)
Context Window
128K
License
Apache 2.0
Release Date
2026-04-02
Benchmark Performance
AA Intelligence Index
—
LMArena Elo
—
HLE
—
ARC-AGI-2
—
SWE-bench Verified
—
GPQA Diamond
58.6
MMLU-Pro
—
LiveCodeBench
52.0
AIME 2025
—
MATH-500
—
API Pricing
API pricing for this model is not yet available