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.
#69 of 89 in Multimodal.
On MIEB lite · Document Understanding · English, EVA02 CLIP B 16 scores 26.2 vs 28.7 — lower.
Released 2023-04-26. 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.
EVA02 CLIP B 16 is an open-weights embeddings & retrieval model from QuanSun, released Apr 2023.
Jina Embeddings v5 Omni Nano (better), Argus Colqwen3.5 9B v0 (open weights).
BetterJina Embeddings v5 Omni NanounstatedOpen×2.8—$0.02DOpenArgus Colqwen3.5 9B v0unstated×3.4——Memory (4-bit): 3 GB → 5.3 GB
BetterEmbeddingGemma 2unstatedOpen×2.6—$0.02
Nemotron Colembed VL 8B v2Open×3.4——Memory (4-bit): 3 GB → 5.2 GBDArgus Colqwen3.5 4B v0unstatedOpen×3.4——vVultronRetrieverPrime Qwen3.5 8BOpen×3.4——Memory (4-bit): 3 GB → 5 GBvVultronRetrieverCore Qwen3.5 4.5BOpen×3.4——