There is no per-unit API price: API price not public.
There is no free API. The weights are open, so you can run it yourself.
#107 of 200 in Embeddings.
On JMTEB v2 · STS · Japanese, Bekko Embedding v1 A25m scores 81.5 vs 81.6 — about the same.
Released 2026-07-28. 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.
Bekko Embedding v1 A25m is an open-weights embeddings & retrieval model from hotchpotch, released Jul 2026.
EmbeddingGemma 300M (better), Ja Embed Sts (open weights).
BetterEmbeddingGemma 300MunstatedOpen+4%—$0.002Max input: 8K tokens → 2K tokens+1NOpenJa Embed Sts+5%——Max input: 8K tokens → 1K tokens+1iBetterE5 Mistral 7B InstructunstatedOpen+3%—$0.022Memory (4-bit): 0.1 GB → 4.3 GBcBetterF2LLM v2 8BunstatedOpen+5%——Memory (4-bit): 0.1 GB → 4.5 GBcBetterF2LLM v2 14BunstatedOpen+4%——Memory (4-bit): 0.1 GB → 8.4 GB
Harrier OSS v1 0.6BunstatedOpen+4%——Memory (4-bit): 0.1 GB → 0.4 GBcF2LLM v2 4BunstatedOpen+4%——Memory (4-bit): 0.1 GB → 2.4 GB