Curiosity · AI Model

Vertex AI textembedding-gecko

textembedding-gecko is Google Cloud's flagship managed text-embedding endpoint on Vertex AI. Built on the Gecko family of small dense encoders distilled from larger Gemini models, it exposes a stable 768-dimensional embedding API optimised for RAG, semantic search, and classification at enterprise scale.

Model specs

Vendor
Google
Family
Gecko embeddings
Released
2023-07
Context window
2,048 tokens
Modalities
text
Input price
$0.025/M tok
Output price
n/a
Pricing as of
2026-04-20

Strengths

  • Managed endpoint with Google Cloud SLA and IAM
  • Integrated with Vertex AI Search and BigQuery
  • Distilled from larger Gemini teacher models

Limitations

  • Closed — you cannot run it outside Google Cloud
  • English-centric — use textembedding-gecko-multilingual for other languages
  • Below research-grade open-weights LLM embedders on peak MTEB scores

Use cases

  • Enterprise RAG over Google Cloud Storage / BigQuery corpora
  • Semantic search in Vertex AI Search applications
  • Classification and clustering pipelines via Dataflow
  • Hybrid search combining BM25 and dense embeddings

Benchmarks

BenchmarkScoreAs of
MTEB English average (Gecko technical report)strong vs. comparable-size encoders2024-03

Frequently asked questions

What is textembedding-gecko?

textembedding-gecko is Google Cloud's managed text-embedding endpoint on Vertex AI. It exposes a stable 768-dimensional embedding API for retrieval, semantic search, and classification.

How is it priced?

Vertex AI charges per 1,000 input characters or per million input tokens (current rates around a few cents per million tokens). Check the Vertex AI pricing page for the latest rates.

How does Gecko compare to NV-Embed or SFR-Embedding?

Gecko is smaller, cheaper, and managed — ideal for enterprise integration on Google Cloud. NV-Embed and SFR-Embedding are larger open-weights models with higher peak retrieval quality.

Sources

  1. Google Cloud — Text embeddings API — accessed 2026-04-20
  2. arXiv — Gecko embeddings paper — accessed 2026-04-20