Model comparison

embeddinggemma-300m-qat-q8_0-unquantized vs jina-embeddings-v5-text-small

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: embeddinggemma-300m-qat-q8_0-unquantized's KV cache at 32K is 24.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

embeddinggemma-300m-qat-q8_0-unquantizedjina-embeddings-v5-text-small
Parameters303M596M
Architecturegemma-embeddingqwen3
Layers2428
Native context2,04832,768
Mixture of expertsnono
Quantizations published156
Smallest quantization0.31 GiB0.19 GiB
Q4_K_M—0.37 GiB
Licencegemmacc-by-nc-4.0

KV cache by context

the term that decides long-context viability
Contextembeddinggemma-300m-qat-q8_0-unquantizedjina-embeddings-v5-text-smallRatio
4,0960.04 GiB0.44 GiB12.44×
8,1920.05 GiB0.88 GiB17.23×
16,3840.08 GiB1.75 GiB21.33×
32,7680.14 GiB3.50 GiB24.22×
65,5360.27 GiB7.00 GiB25.97×
131,0720.52 GiB14.00 GiB26.95×