On November 12, 2026, Cloudflare announced in its Workers AI changelog that EmbeddingGemma 2 (@cf/google/embeddinggemma-2) is available on Workers AI. This release covers the model's availability, specifications, and pricing on Cloudflare's Workers AI platform.

Shared 768-D Multimodal Space, Flexible Dimensions & Usage

Described by Cloudflare as an open multimodal embedding model from Google DeepMind, EmbeddingGemma 2 maps text (including code), images, video, and audio into a single 768-dimensional vector space so a single text query can retrieve mixed products, documents, and media:

Multimodal Inputs, 8,192-Token Context & Code Support: Combines multiple modalities in a single request sharing an 8,192-token context window, with support for over 100 languages and improved code retrieval over the original EmbeddingGemma.

Flexible Output Dimensions & Task Prefixes: Supports setting dimensions to 128, 256, or 512 (instead of the full 768) to reduce storage footprint with a small cost to quality, and accepts task prefixes (such as task: search result | query: ...) for search, classification, clustering, or similarity.

Invocation & Zero Data Retention: Text queries can use the input field, while images, video, and audio are passed as base64 data URIs inside messages via env.AI.run("@cf/google/embeddinggemma-2", { messages: ... }). The model info card lists Zero data retention as Yes.

Pricing & Availability Scope

Per-token USD and Neuron pricing: Across the changelog, model info card, and platform pricing table, @cf/google/embeddinggemma-2 is listed at $0.0017 per million input tokens, with the pricing table also specifying 154 neurons per million input tokens. Availability applies specifically to Workers AI.