AI Index

EmbeddingGemma 2

Embedding Google Partly verified Released Oct 6, 2026
  • In TextImageAudioVideo
  • Open weights

Prices

0 offerings · prices as listed

No priced offering is listed for this model yet.

Details

Shortest vector
128
Languages
100

Not stated on Google’s pages: vector types, inputs per request.

Run locally

Weights: google/embeddinggemma-2
  • Parameters 0.7B
  • Dense
  • Native precision BF16
  • Layers 24
  • KV cache per token 48 KB
  • Config positions 256K (config.json max_position_embeddings)
My hardware GeForce RTX 4090 24 GB · 1008 GB/s · 64 GB DDR5 dual channel · 89.6 GB/s
On your hardware · GeForce RTX 4090
Runs well BF16 (native) (automatic) 8K context 320–479 tok/sOne GPU
WeightsKV cacheOverheadUsable
Needs
≈ 2.0 GB (weights 1.4 (maker’s file) + KV cache 0.1 + overhead 0.5)
Usable
21.6 GB (90% of its memory)
Generation speed
320–479 tok/s
Longest context
256K (the model’s limit)
First token
about 0.02–0.04 s for a 2K prompt
PrecisionWeightsNeeds at 8KOn your hardwareLongest context
BF16 (native)maker’s file1.4 GB≈ 2.0Runs well 320–479 tok/s256K
Q8_00.7 GB≈ 1.3Runs well 573–860 tok/s256K
FP80.7 GB≈ 1.3Runs well 605–908 tok/s256K
Q6_K0.6 GB≈ 1.2Runs well 721–1081 tok/s256K
Q5_K_M0.5 GB≈ 1.1Runs well 816–1224 tok/s256K
Q5_00.5 GB≈ 1.1Runs well 840–1260 tok/s256K
Q4_K_M0.4 GB≈ 1.0Runs well 933–1399 tok/s256K
Q4_00.4 GB≈ 1.0Runs well 994–1492 tok/s256K
IQ4_XS0.4 GB≈ 1.0Runs well 1042–1564 tok/s256K
Q3_K_Mquality drops0.3 GB≈ 0.9Runs well 1115–1673 tok/s256K
IQ3_XXSquality drops0.3 GB≈ 0.8Runs well 1352–2028 tok/s256K
Q2_Kquality drops0.3 GB≈ 0.8Runs well 1372–2059 tok/s256K
IQ2_XXSquality drops0.2 GB≈ 0.8Runs well 1802–2702 tok/s256K

Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.

Compared with other hardware

Each at the highest precision it runs. Click one for details.

Smallest that runs it (Q4_K_M, 8K): GPU GeForce RTX 3080; unified memory Apple M5 · 16 GB.

Run commands

vLLM
vllm serve google/embeddinggemma-2 --max-model-len 8192
Memory to fine-tune
Full≈ 13.7 GBFits your hardware
LoRA≈ 4.0 GBFits your hardware
QLoRA≈ 3.0 GBFits your hardware

Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.

Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated

Cited scores

From third-party leaderboards, not measured by us

No cited leaderboard covers embedding models yet. MTEB’s results are open (CC0); we will cite them per benchmark, never as a combined score.

News

All news →
  • Nov 2, 2026Release Google releases Gemini 4 Argon and announces Gemini 4 Carbon Google rolled out Gemini 4 Argon globally across Google AI Studio, the Gemini API, Google Cloud, Search and the Gemini app. Gemini 4 Carbon is in pre-release testing with the U.S. government and selected cyber defenders, with a broader launch planned.
  • Oct 6, 2026Release Google releases EmbeddingGemma 2 Google launched EmbeddingGemma 2, a 740-million-parameter multimodal embedding model for text, code, images, video and audio, under the Apache 2.0 license. It has an 8K-token context window and is available to download on Hugging Face and Kaggle.

Changes

All changes →
  1. New embedding model

Sources

Google’s official policies: data use, privacy, terms →

Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data

Google · United States · ai.google.dev

Snapshots, the same family and other models from this provider