There is no per-unit API price: self-host only.
There is no free API. The weights are open, so you can run it yourself.
#131 of 200 in Embeddings.
On MTEB Dutch v1 · Classification · Dutch, KaLM Embedding Multilingual Mini Instruct v1 scores 60.5 vs 60.6 — about the same.
Released 2024-10-23. It is the current version in its line.
Yes, the weights are open (MIT licence).
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
KaLM Embedding Multilingual Mini Instruct v1 is an open-weights embeddings & retrieval model from HIT-TMG, released Oct 2024.
EmbeddingGemma 300M (better), F2LLM v2 8B (open weights).
BetterEmbeddingGemma 300MunstatedOpen+6%—$0.002Max input: 33K tokens → 2K tokenscOpenF2LLM v2 8Bunstated+23%——Memory (4-bit): 0.3 GB → 4.5 GB
BetterQwen 3 Embedding 8BunstatedOpen+21%—$0.01Memory (4-bit): 0.3 GB → 4.5 GBiBetterMultilingual E5 Large InstructunstatedOpen+7%—$0.01Max input: 33K tokens → 1K tokens
Jina Embeddings v3unstatedOpen+3%—$0.05Max input: 33K tokens → 8K tokens
Llama Embed Nemotron 8BOpen+21%——Memory (4-bit): 0.3 GB → 4.5 GBcF2LLM v2 4BunstatedOpen+21%——Memory (4-bit): 0.3 GB → 2.4 GB