EmbeddingGemma 2 for video search: Sume has no embeddings endpoint

The AI Brief of October 7, 2026 reports EmbeddingGemma 2 as a 740M-parameter, Apache 2.0 model that embeds text, code, images, video and audio. People who build a media library search ask whether Sume hosts it.

It does not. As of 2026-10-07 I found no public embeddings endpoint in the Sume docs or API contract, and no EmbeddingGemma entry in the model registry. If you need vectors, you run the model yourself or use an embeddings provider, and keep the vectors in your own store.

What Sume sells for finding things instead

These are the priced, documented jobs that answer parts of a search problem without vectors. Prices are the published SKUs read 2026-10-07.

A split that works

Embed in your own stack, generate in Sume. Transcripts from the $0.01 per minute STT job are plain text, so any embedder, EmbeddingGemma 2 included, can index them. Ten hours of audio is 600 minutes, which is $6.00 of transcription.