All tasks / Models

KaLM-Embedding-Gemma3-12B-2511

Tencent's open-weights 12B multilingual text embedding model built on Gemma 3.

Tencent
Open weights
Released Nov 2025

Where it ranks

S
Embeddings & retrieval
Models · Local · 72.3% on MTEB(Multilingual, v2)

Facts

Vendor: Tencent
Family: KaLM-Embedding
Released: Nov 2025
Weights: Open: download and run it yourself
License: Tencent KaLM Embedding Community License
Parameters: 11.77B
GPU memory to run locally: ~24 GB (Derived: 11.77B parameters x 2 bytes = ~23.5 GB of bf16 weights (the board's 44.9 GB figure is fp32).)
Access: Free (open weights)
Last verified: 19 Oct 2026

Benchmark results

Benchmark	Result	Entry on the leaderboard	As of	Source
MTEB(Multilingual, v2)	72.3%	`tencent/KaLM-Embedding-Gemma3-12B-2511` Mean (Task) over all 131 tasks; board rank 2 (Borda); board zero-shot 73%; read from the leaderboard's public API.	19 Oct 2026	mteb-leaderboard-backend.hf.space

Where you can use it

No tracked tool lists this model yet; as open weights it runs in local runtimes.

Compare

Alternatives to KaLM-Embedding-Gemma3-12B-2511

Sources