# Retrieve across languages

> Retrieve across languages: 0 companies fully do it, 12 with a limit, 35 carry the tag. Every listing in Embeddings & rerankers with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/embeddings.multilingual-retrieval/
- Markdown: https://www.anchorterminal.com/niches/embeddings.multilingual-retrieval/index.md (~1,750 tokens)
- Slim: https://www.anchorterminal.com/niches/embeddings.multilingual-retrieval/index.min.md (~530 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/embeddings.multilingual-retrieval/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 0 | 12 | 35 | 4 | 4 |

## What counts as done

The listing names the languages it supports, documents cross-language retrieval, and says what language task instructions or prefixes should be written in; distinct words in non-Latin scripts get distinct vectors.

Measured, for our own tests: recall@10 on a labelled set with queries in one language and answers in another.

How agents ask for it: search our Swedish documents; query in English and find the Spanish policy; RAG over Hindi and English text; does this model support Arabic

## Does the job, with a limit

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.

- **Cohere Embed and Rerank (Cohere Inc.)**: A table of over 100 supported languages, and task setting is an enum so no instruction language applies, but cross-language retrieval isn't documented.
- **EmbeddingGemma 2 (Google LLC (RDAP registrant of gemma.ai; the card names Google DeepMind as author))**: Claims 100+ languages and cross-lingual alignment but gives no language list or instruction language. An account measured 0.7279 NDCG@10 on Hindi IndicQA from the text path.
- **EmbeddingGemma 300M (Google LLC (RDAP registrant of gemma.ai; the card names Google DeepMind as author))**: Trained on 100+ languages with MTEB(Multilingual) scores, but no language list, no cross-language retrieval statement and no instruction-language guidance.
- **Gemini Embedding (Google LLC)**: Claims over 100 languages but gives no language list, no cross-language retrieval statement and no prefix-language guidance.
- **Jina Embeddings and Reranker (Jina AI GmbH)**: Text embeddings are described as covering 100+ languages, but there is no language list or cross-language retrieval statement in the reference.
- **Nomic Embed (Nomic, Inc.)**: Atlas has gte-multilingual-base for over 70 languages without naming them, and an account found nomic-embed-text v1.5 in Ollama gave identical vectors for different Chinese words.
- **Qwen3-Embedding-0.6B (Alibaba Cloud (the registrant of qwen.ai per RDAP, carried from the Qwen3.8-27B check))**: Over 100 languages and cross-lingual retrieval are stated, but no language list, and the card advises writing instructions in English even for other languages, a named blocker.
- **Qwen3-Embedding-4B (Alibaba Cloud (named in the copyright line of the 8B sibling's Apache 2.0 licence; this repository has no LICENSE file))**: Over 100 languages and cross-lingual retrieval are stated, but no language list, and the card advises writing instructions in English even for other languages, a named blocker.
- **Qwen3-Embedding-8B (Alibaba Cloud (named in the licence's copyright line))**: Over 100 languages and cross-lingual retrieval are stated, but no language list, and the card advises writing instructions in English even for other languages, a named blocker.
- **Qwen3-VL-Embedding-2B (Alibaba Cloud (RDAP names Alibaba Cloud (Singapore) Private Limited as the registrant of qwen.ai))**: Claims 30+ languages, inherited from Qwen3-VL, but names none and documents no cross-language retrieval; the series advises English instructions.
- **Qwen3-VL-Embedding-8B (Alibaba Group (Tongyi Lab, named in the copyright line of the GitHub repository's Apache 2.0 licence))**: Claims 30+ languages, inherited from Qwen3-VL, but names none and documents no cross-language retrieval; the series advises English instructions.
- **Voyage AI embeddings and rerankers (Voyage AI Innovations, Inc.)**: Models are described as optimised for multilingual retrieval, and an account embedded Swedish text, but the docs give no language list or cross-language retrieval statement.
- **ZeroEntropy zerank and zembed (ZeroEntropy, Inc.)**: zembed-1 is tagged and described as multilingual, but names no languages and documents no cross-language retrieval.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **NVIDIA NeMo Retriever Embedding and Reranking NIMs (NVIDIA Corporation)**: The support matrix and API page name no languages; per-model cards on NGC or Hugging Face weren't read.

- not this job: Amazon Nova Multimodal Embeddings (Amazon Web Services, Inc.): Neither the guide, the schema nor the Bedrock model card names a supported language or documents cross-language retrieval.
- not this job: Granite Embedding Small English R2 (International Business Machines Corporation): English only, as the card states.
- not this job: jina-embeddings-v5-omni-small (Jina AI GmbH, a subsidiary of Elastic N.V. (per the jina-embeddings listing's 2 October 2026 check)): Tagged multilingual but names no languages and documents no cross-language retrieval, a blocker the map names for this model.
- not this job: Mistral Embed and Codestral Embed (Mistral AI (RCS Paris 952 418 325)): No language list for either embedding model and no cross-language retrieval statement, a blocker the map names for Mistral.
- not this job: OpenAI embeddings (OpenAI OpCo, LLC): The guide and reference name no supported languages and document no cross-language retrieval.

## What stops agents finishing

- no language list for the model (Mistral, jina-embeddings-v5-omni-small)
- tokeniser maps different non-Latin words to the same vector (nomic-embed-text v1.5 in Ollama)
- instructions must be in English even for other languages (Qwen3)

## Around it

Before: translation.translate-text. After: vector-search.rag-over-docs.

## Where this comes from

Weight 4 of 100 in Embeddings & rerankers: 4 of 62 accounts (91 account-use-case tags, accounts counted in each use case they fit, scaled to 100 and rounded by hand); Swedish, Hindi, Chinese and enterprise cross-language accounts; thin evidence

As data: https://www.anchorterminal.com/api/v1/use-cases/embeddings.multilingual-retrieval.json, or the MCP tool get_use_case with id embeddings.multilingual-retrieval. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
