Niches · Embeddings & rerankers
Classify, cluster, deduplicate or cache by similarity
Four or more companies' docs show a full fit.
What counts as done
The model documents a classification or clustering mode (or says one vector serves both), the price per million tokens or the local cost lets the whole set be embedded, and the docs say whether similarity scores are comparable across calls and models.
Measured, for our own tests: accuracy of a linear classifier trained on a labelled sample of the vectors. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: group these tickets by topic
, label these emails
, find near-duplicate records
, has this question been answered already
Does the job 13
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| Cohere Embed and Rerank BB Cohere Inc. | classification and clustering are documented input_type values, Pro and Fast share a space, and accounts use it daily for clustering and for categorising RSS feeds at negligible cost. Classifier accuracy is unmeasured. source |
| EmbeddingGemma 2 B Google LLC (RDAP registrant of gemma.ai; the card names Google DeepMind as author) | Classification and Clustering prompt names are documented as symmetric tasks, it runs locally at hardware cost, and the card says embeddings compare directly across modalities. Classifier accuracy unmeasured. source |
| EmbeddingGemma 300M D Google LLC (RDAP registrant of gemma.ai; the card names Google DeepMind as author) | Classification and Clustering prompts are documented, it runs locally, and an account trains RBF SVMs on its vectors with good separation. Linear-classifier accuracy unmeasured. source |
| Gemini Embedding B Google LLC | task: classification and task: clustering prefixes are documented, prices are published with a free tier, and the guide says 001 and 2 vectors can't be compared. Classifier accuracy unmeasured. source |
| Jina Embeddings and Reranker B Jina AI GmbH | classification and clustering are task values on v5, and prices per token are in the model catalogue. Classifier accuracy unmeasured. source |
| jina-embeddings-v5-omni-small D Jina AI GmbH, a subsidiary of Elastic N.V. (per the jina-embeddings listing's 2 October 2026 check) | Classification and clustering adapters are selected with default_task, with pre-merged variants for each, and it runs locally at hardware cost for non-commercial use. Classifier accuracy unmeasured. source |
| Nomic Embed E Nomic, Inc. | classification (for linear probes) and clustering (high linear separability) are documented task types, and local inference is free. Classifier accuracy unmeasured. source |
| OpenAI embeddings BB OpenAI OpCo, LLC | The guide presents one vector for clustering and classification with worked examples, pricing is per input token, and accounts use text-embedding-3-small as an email classifier input and as a cosine-0.85 answer cache. Classifier accuracy unmeasured. source |
| Qwen3-Embedding-0.6B B Alibaba Cloud (the registrant of qwen.ai per RDAP, carried from the Qwen3.8-27B check) | Instruction-aware, so the task is set by a one-sentence instruction, with MTEB classification and clustering columns published, and it runs locally at hardware cost. Classifier accuracy unmeasured. source |
| Qwen3-Embedding-4B BB Alibaba Cloud (named in the copyright line of the 8B sibling's Apache 2.0 licence; this repository has no LICENSE file) | Instruction-aware, so the task is set by a one-sentence instruction, with MTEB classification and clustering columns published, and it runs locally at hardware cost. Classifier accuracy unmeasured. source |
| Qwen3-Embedding-8B BB Alibaba Cloud (named in the licence's copyright line) | Instruction-aware, so the task is set by a one-sentence instruction, with MTEB classification and clustering columns published, and it runs locally at hardware cost. Classifier accuracy unmeasured. source |
| Qwen3-VL-Embedding-2B C Alibaba Cloud (RDAP names Alibaba Cloud (Singapore) Private Limited as the registrant of qwen.ai) | Instruction-aware, with MMTEB classification and clustering columns and MMEB-V2 image and video classification results published, and it runs locally at hardware cost. Classifier accuracy unmeasured. source |
| Qwen3-VL-Embedding-8B B Alibaba Group (Tongyi Lab, named in the copyright line of the GitHub repository's Apache 2.0 licence) | Instruction-aware, with MMTEB classification and clustering columns and MMEB-V2 image and video classification results published, and it runs locally at hardware cost. Classifier accuracy unmeasured. source |
Does the job, with a limit 5
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Amazon Nova Multimodal Embeddings BB Amazon Web Services, Inc. | CLASSIFICATION and CLUSTERING are documented embeddingPurpose values, but the docs don't say whether scores compare across purposes, dimensions or calls. source |
| Granite Embedding Small English R2 E International Business Machines Corporation | Runs locally at hardware cost and reports an MTEB-v2 average over 41 tasks, but documents no classification or clustering mode and doesn't say one vector serves all tasks. source |
| Mistral Embed and Codestral Embed B Mistral AI (RCS Paris 952 418 325) | The guides present one vector for clustering and classification with worked examples, but don't say whether scores compare across calls or between the two models. source |
| NVIDIA NeMo Retriever Embedding and Reranking NIMs B NVIDIA Corporation | The overview claims classification and clustering, but the API page documents only query and passage modes, and says nothing on comparing scores across models. source |
| Voyage AI embeddings and rerankers B Voyage AI Innovations, Inc. | input_type null gives plain vectors and is documented as compatible with query and document vectors, and prices are published, but no classification or clustering mode exists on current models; the FAQ's instructions apply to voyage-large-2-instruct. source |
Doesn't do this job 1
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| ZeroEntropy zerank and zembed F ZeroEntropy, Inc. | The hosted API is gone, and the zembed-1 card documents only query and document retrieval, with no classification or clustering mode. source |
What stops agents finishing
- no clustering or classification mode and no guidance on which input type to use
- similarity thresholds tuned on one model do not carry to another
Around it
Before: data.export-table, email.list-messages. After: decision-models.classify.
Where this comes from
Weight 8 of 100 in Embeddings & rerankers: 7 of 62 accounts (91 account-use-case tags, accounts counted in each use case they fit, scaled to 100 and rounded by hand); email, RSS and complaint clustering, semantic caching and classifier training on embeddings
The same as data: /api/v1/use-cases/embeddings.classify-cluster.json, or the MCP tool get_use_case with id embeddings.classify-cluster. Other jobs in this category: /api/v1/niches?category=embeddings.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.