A transparent planning tool, not a vendor quote.
This is an independent calculator and is not affiliated with Google. It runs in your browser and does not upload prompts, media, files or credentials.
Why it exists
EmbeddingGemma 2 combines text, code, images, video and audio in one embedding space. The official documentation describes the model well, but developers still have to combine several separate facts when deciding whether an input fits and how large an index may become. This tool puts those decisions in one place.
Input-budget calculation
The estimate adds text tokens, image tokens, sampled video-frame tokens and audio tokens. The default assumptions use 280 tokens per image or video frame, one video frame per second and 25 tokens per second of audio. All modalities share the 8,192-token context window.
Vector-storage calculation
Raw vector bytes are calculated as vector count × dimensions × bytes per value. Metadata, configurable index overhead and replicas are then added. Provider compute, operations, backups, minimum capacity and network charges are not included.
Model-footprint calculation
The model estimate multiplies loaded encoder parameters by two bytes for BF16 weights. It is not a complete VRAM or RAM estimate; frameworks, activations, caches and media processing require additional memory.
Primary sources
Last reviewed
Sources and default assumptions were last reviewed on October 12, 2026. Always verify current documentation and benchmark your own retrieval data before making a production decision.