Skip to content
Local AI Wizard
Your machine
Theme

EmbeddingGemma (300M)

Google · google/embeddinggemma-300m

Weights license: Publisher terms: Gemma Terms of Use · code and components may use other terms

MTEB (Multilingual, v2)61.1 / 100mean(task) · independent run
Memory needed~3 GB8-bit planning estimate · not measured
Parameters0.3 BSee technical details for the basis
Dimensions · context7682,048 tokens · Matryoshka 512/256/128

Quality

MTEB (Multilingual, v2)

Mean(task) · 0–100 · higher is better · this model outlined

  • Color = publisher

Source: MTEB results dataset (independent runs, CC0) · publisher-reported where the model card is the only source · other benchmarks and versions are not mixed in

Sourced benchmarks
MTEB (Multilingual, v2)mean(task) · 768d61.14independent run
2026-10-25
MTEB (eng, v2)mean(task) · 768d69.67independent run
2026-10-25
MTEB (Multilingual, v2)mean(task) · 768d61.15publisher-reported
2026-10-07
MTEB (eng, v2)mean(task) · 768d69.67publisher-reported
2026-10-07
MTEB (Code, v1)mean(task) · 768d68.76publisher-reported
2026-10-07

Reported by the publisher

6 results · Google's model card · read 2026-10-25
Mean (Task)
61.15MTEB (Multilingual, v2)
Mean (TaskType)
54.31MTEB (Multilingual, v2)
Mean (Task)
69.67MTEB (English, v2)
Mean (TaskType)
65.11MTEB (English, v2)
Mean (Task)
68.76MTEB (Code, v1)
Mean (TaskType)
68.76MTEB (Code, v1)

Values as printed on the card, with the publisher's own settings (card section: Evaluation › Benchmark Results › Full Precision Checkpoint); column “768d”. Not comparable across publishers and not used in any ranking. Model card

Quality × memory

No comparable quality × memory pair

Embedding memory estimates cluster at a few GB, so a quality × memory chart would not separate these models; MTEB is ranked above instead.

Consider it for

Consider EmbeddingGemma (300M) for retrieval and semantic search: it turns text into 768-dimension vectors.

Check first

The publisher's memory figures are for specific setups and are under How to run, with the prompt format; the hardware list uses the planning estimate. PT-BR is confirmed by a source.

Runs on

Fast memory holds it 119

The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.

Show all 119

Hardware profiles grouped by the embedding planning estimate (8-bit weights, 20% allowance, 2 GB reserve). Not a measured fit.

Measured speed

No published measurement for this checkpoint

Throughput (texts per second) is shown only when a source publishes it with hardware, build and setup. It is never estimated.

How to run

Published memory: 200 MB (“Less than 200MB of RAM with quantization” (upper bound, Google blog, 2025-09-04); no device or runtime named.). The planning estimate uses 8-bit weights, 20% working allowance and a 2 GB reserve; it is a ceiling, not a measurement.

Runs in
Ollama, Sentence Transformerssource
Query/document prompts
Query prompts: 'task: {task description} | query: ' (default task description 'search result'). Document prompts: 'title: {title | "none"} | text: '. Prompts are prepended to the input strings.
Languages · PT-BR
100+ languages (Google model card and blog)PT-BR confirmed by a source

Documented ways to run it

The model card could not be read (missing or gated)

Choose my computer

Technical details

What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.

Checkpoint / source
google/embeddinggemma-300m
Task
embeddings for search (feature extraction)
Library in source
Ollama, Sentence Transformers
Weight license
Publisher terms: Gemma Terms of Use
Commercial use
Allowed Gemma Terms of Use (last modified 2025-03-24; verbatim copy in the ScanCode LicenseDB, homepage ai.google.dev/gemma/terms) · read 2026-10-18license text
Parameter basis
Hub safetensors total (302,863,104 parameters, read 2026-10-07). Google's blog says 308M and the MTEB metadata 307.6M; the sources round differently.
Announced
2025-09-04

The Hub repository is gated (licence acceptance), so its card cannot be read anonymously. Facts here come from Google's model-card page on ai.google.dev, its blog, the Hub API and the MTEB results dataset. Google's own MTEB numbers (61.15 Multilingual v2, 69.67 English v2, 68.76 Code v1 at 768 dimensions) match the independent run for the first two. Google's pages do not name Portuguese; the MTEB metadata lists 19 training languages without Portuguese. PT-BR is confirmed here only through the Portuguese-only MTEB subsets, which is a small indicator.

Sources

Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.

Curated facts checked: 2026-10-07. Each value cites its source (the publisher's card or the MTEB results dataset); estimates are labelled as estimates, and a value no source states stays empty. A repository commit is not a model release date.