Is hs-embed-1 state of the art?Pending benchmarks.Announced Nov 3, 2026; no independent results yet.
Vectorize's 112M multilingual text embedding model for Hindsight memory recall: 384-dimension vectors (truncatable to 64) over inputs up to 8,192 tokens in eight languages. Open weights (MIT) on Hugging Face, also served on the Hindsight Inference API.
The details so far
- Lab
- Vectorize
- Announced
- Nov 3, 2026
- API
- Live · hs-embed-1 · Hindsight Inference POST /v1/embeddings (OpenAI Embeddings compatible), $0.005 per 1M input tokens
- Open weights
- Published · Hugging Face hindsight-ai/hs-embed-1: FP16 safetensors (FP32 on an fp32 branch) and an int8 ONNX export; serves on Text Embeddings Inference on CPU or GPU
- Parameters
- 112M (111,728,256 stored parameters; 98M are the token embedding table, so each pass runs a 13M-parameter transformer)
- Type
- Multilingual text embedding model; mean pooling, cosine similarity, no query or document prompts
- Base model
- hotchpotch/bekko-embedding-v1-a25m (fine-tuned)
- Dimensions
- 384, truncatable to 256, 128 or 64
- Context
- 8,192 tokens
- Languages
- English, Spanish, French, German, Portuguese, Japanese, Korean, and Simplified and Traditional Chinese
- License
- MIT
- Siblings
- hs-decide-1-flash and hs-rerank-1, released the same day (separate pages)
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: overall mean of 56.8 over twelve retrieval benchmarks (including NanoBEIR, MIRACL, BEAM, LongEmbed and LMEB) against 54.6 for granite-embedding-97m-multilingual-r2 and 50.3 for multilingual-e5-small.
Lab claim: 16.5 ms median query latency on two CPU threads.
Sources
LABIntroducing Hindsight Models (Hindsight blog)hindsight.vectorize.io/blog/2026/11/03/hindsight-models · Nov 3, 2026LABVectorize announcement on Xx.com/Vectorizeio/status/2117647994048704588 · Nov 3, 2026LABhs-embed-1 benchmarks (Hindsight Benchmarks)benchmarks.hindsight.vectorize.io/models/embeddingsREPOhindsight-ai/hs-embed-1 on Hugging Facehuggingface.co/hindsight-ai/hs-embed-1Tracking continues until an independent board lists it.