KaLM Embedding Gemma3 12B 2511

Nov 2025
KaLM Embedding Gemma3 12B 2511
Compare in
on
self-host onlyprice
82.7MTEB Scandinavian v1
Nov 2025released
3,840Dimensions
131K tokensMax input

Ranking?

Full ranking →
1Gemini Embedding 2 Preview95.82Jina Embeddings v5 Omni Small862Jina Embeddings v5 Text Small864Jina Embeddings v5 Omni Nano864Jina Embeddings v5 Text Nano86
···
8Qwen 3 Embedding 8B84.79KaLM Embedding Gemma3 12B 251182.710Qwen 3 Embedding 4B82.2

Run it yourself

estimate
Fits on✓ 8 GB (phone, laptop) · compressed✓ 16 GB · 8-bit✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
Licencetencent-kalm-embedding-communitycommercial use with conditions

Benchmarks

in Embeddings & Retrieval
MTEB Europe v1 · Bitext Mining · multilingual (28 languages)92.8#5
MTEB Europe v1 · Clustering · multilingual (29 languages)50.7#8
MTEB Europe v1 · Instruction Reranking · English5.5#9
MTEB Europe v1 · Reranking · multilingual (12 languages)63.1#8
MTEB Europe v1 · Multilabel Classification · multilingual (23 languages)25.2#7
MTEB French v1 · Pair Classification · French69.2#12
MTEB Indic v1 · Reranking · multilingual (2 languages)87.6#15
MTEB Indic v1 · Pair Classification · Hindi82.5#8
MTEB Indic v1 · STS · multilingual (13 languages)63.4#2
MTEB Indic v1 · Clustering · multilingual (21 languages)52.9#14
MTEB Dutch v1 · Reranking · Dutch90.5#28
MTEB Portuguese v1 · Multilabel Classification · Portuguese31.6#4
FollowIR · Mean over tasks · English5.5#9
FollowIR · Mean over task types · English5.5#9
FollowIR · Instruction Reranking · English5.5#9
MTEB Spanish v1 · Pair Classification · Spanish77.7#8
MTEB Thai v1 · Mean over tasks · multilingual (2 languages)66.4#7
MTEB Thai v1 · Mean over task types · multilingual (2 languages)66.8#8
MTEB Thai v1 · Clustering · Thai57.2#18
MTEB Thai v1 · Classification · Thai76.2#9
MTEB Thai v1 · Pair Classification · Thai85.1#7
MTEB Thai v1 · Retrieval · multilingual (2 languages)63.7#11
MTEB Thai v1 · Reranking · Thai51.7#21
MTEB Scandinavian v1 · Bitext Mining · multilingual (3 languages)82.7#9
MTEB Russian v1.1 · Pair Classification · Russian61.6#27
MTEB multilingual v2 · Mean over tasks · multilingual (1037 languages)72.3#2
MTEB multilingual v2 · Mean over task types · multilingual (1037 languages)62.5#2
MTEB multilingual v2 · Classification · multilingual (119 languages)77.9#2
MTEB multilingual v2 · Clustering · multilingual (208 languages)55.8#11
MTEB multilingual v2 · STS · multilingual (35 languages)79#7
MTEB multilingual v2 · Pair Classification · multilingual (24 languages)84.7#7
MTEB multilingual v2 · Retrieval · multilingual (120 languages)75.7#3
MTEB multilingual v2 · Bitext Mining · multilingual (1026 languages)83.8#3
MTEB multilingual v2 · Instruction Reranking · English5.5#9
MTEB multilingual v2 · Multilabel Classification · multilingual (25 languages)33#8
MTEB multilingual v2 · Reranking · multilingual (22 languages)67.3#10

FAQ

How much does KaLM Embedding Gemma3 12B 2511 cost?›

There is no per-unit API price: self-host only.

Is KaLM Embedding Gemma3 12B 2511 free to use?›

There is no free API. The weights are open, so you can run it yourself.

How good is KaLM Embedding Gemma3 12B 2511, and what is it best at?›

#6 of 205 in Embeddings.

Is KaLM Embedding Gemma3 12B 2511 better than Qwen 3 Embedding 8B?›

On MTEB Scandinavian v1 · Bitext Mining · multilingual (3 languages), KaLM Embedding Gemma3 12B 2511 scores 82.7 vs 84.7 — lower.

When was KaLM Embedding Gemma3 12B 2511 released? Is there a newer version?›

Released 2025-11-06. It is the current version in its line.

Is KaLM Embedding Gemma3 12B 2511 open source?›

Yes, the weights are open (tencent-kalm-embedding-community licence).

Can I run KaLM Embedding Gemma3 12B 2511 locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

What is KaLM Embedding Gemma3 12B 2511?›

KaLM Embedding Gemma3 12B 2511 is an open-weights embeddings & retrieval model from Tencent, released Nov 2025.

What are the best alternatives to KaLM Embedding Gemma3 12B 2511?›

Jina Embeddings v5 Text Nano (better), Jina Embeddings v5 Text Small (open weights).