EmbeddingGemma 300M

Overview

Google’s EmbeddingGemma 300M generates 768-dimension text embeddings from a compact 300M-parameter architecture. It is designed specifically for bulk corpus indexing, semantic deduplication, and retrieval-augmented generation pipelines where cost efficiency and throughput matter more than maximum representational depth. On Kyma, the model runs through an OpenAI-compatible endpoint with automatic request failover and exact cost reporting in the usage object. It supports prompt caching, which bills repeated prefixes at this model’s cached input rate. The model accepts text input up to a 2048-token context window and outputs fixed-length vectors. It does not support text generation, reasoning, vision, or structured outputs, and it is strictly an embedding-only endpoint.

Specs

Field	Value
Model ID	`embeddinggemma-300m`
Best for	Bulk corpus indexing, RAG retrieval on a budget, semantic dedup
Context window	2,048
Input modalities	Text
Output modalities	Text
Tool calling	Yes
Structured outputs	Yes
Prompt caching	Yes
Speed	Fast
Cost band	Cheap
Release stage	Stable

Pricing

	Per 1M tokens
Input	$0.0027
Output	$0.00

Use this when

Bulk Document Indexing — Index large text corporas efficiently while keeping per-token embedding costs minimal.
Budget RAG Retrieval — Power semantic search in retrieval pipelines where high throughput outweighs the need for larger vector dimensions.
Semantic Text Deduplication — Identify and remove duplicate or near-duplicate entries across large datasets using fast vector similarity.
High-Volume Data Processing — Process massive text batches quickly with a lightweight model optimized for speed and low latency.

Not ideal for

Do not use this model for documents exceeding 2048 tokens, tasks requiring generative text or reasoning, or applications that demand high-dimensional embeddings for fine-grained semantic discrimination.

Pick something else when

You need higher-dimensional embeddings for complex semantic tasks: use qwen3-embedding-8b.

Example

from openai import OpenAI

client = OpenAI(base_url="https://kymaapi.com/v1", api_key="kyma-...")

response = client.embeddings.create(
    model="embeddinggemma-300m",
    input=["first document", "second document"],
)

Agent query example

Ask the API which models fit, instead of hardcoding an id:

curl "https://kymaapi.com/v1/models?recommended_for=search-documents&quality_tier=strong"

FAQ

What is the maximum input length for this model? The context window is capped at 2048 tokens per request. Longer inputs must be chunked before embedding. Does Kyma support prompt caching for this endpoint? Yes. Repeated prompt prefixes are billed at this model’s cached input rate, reducing costs for batch processing with shared headers or instructions. Can I use this model to generate text or answer questions? No. This is an embedding-only model that outputs fixed-length vectors. It does not support text generation, reasoning, or vision tasks.

​Overview

​Specs

​Pricing

​Use this when

​Not ideal for

​Pick something else when

​Example

​Agent query example

​FAQ