Generation-3 multilingual text embedding from Qwen.
text-embedding-v3 is Qwen's generation-3 general-purpose multilingual text embedding model. It produces dense vectors for retrieval, semantic search, clustering, classification, and RAG indexing across Chinese, English, and other major languages.
If you are starting a new project, prefer text-embedding-v4 — v3 is kept here for compatibility with existing indices and pipelines built before the v4 release. TheRouter routes between bailian-cn and bailian-sg per request based on cost.
| Type | Rate |
|---|---|
| Input | $0.0756 / 1M tokens |
| Model family | Alibaba Cloud text-embedding v3help.aliyun.com ↗ | verified |
| Embedding dimensions | 64, 128, 256, 512, 768, or 1,024 (default)help.aliyun.com ↗ | verified |
| Supported languages | Chinese, English, Spanish, French, Portuguese, Indonesian, Japanese, Korean, German, Russian, and 50+ other major languageshelp.aliyun.com ↗ | verified |
| TheRouter context length | 8,192 tokenstherouter.ai ↗ | verified |
| TheRouter input price | $0.144 per 1M input tokenstherouter.ai ↗ | verified |
| Alibaba Cloud public-cloud API compatibility | OpenAI-compatible /v1/embeddings and DashScope native APIshelp.aliyun.com ↗ | verified |
| Primary endpoint on TheRouter | POST /v1/embeddingstherouter.ai ↗ | verified |
| Successor model | qwen/text-embedding-v4qwenlm.github.io ↗ | verified |
| Parameter count | Not publicly disclosed in Alibaba's Model Studio v3 docs | unknown |
| Training cutoff | Not publicly disclosed | unknown |
| License | Not publicly disclosed for the hosted API model | unknown |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Alibaba Cloud documented embedding tasks Alibaba documents the model for semantic search, recommendation, clustering, and classification but does not publish a benchmark table for text-embedding-v3 on the API page. | — | Not publicly disclosed | — |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "qwen/text-embedding-v3",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Generate dense vectors through TheRouter's OpenAI-compatible /v1/embeddings endpoint. Use dimensions to trade recall quality against vector storage and ANN index size.
curl https://api.therouter.ai/v1/embeddings \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen/text-embedding-v3",
"input": [
"TheRouter routes embedding requests through one OpenAI-compatible API.",
"Qwen text-embedding-v3 supports configurable vector dimensions."
],
"encoding_format": "float",
"dimensions": 1024
}'Alibaba Qwen introduced the Qwen3 Embedding and Reranker series with stronger multilingual retrieval positioning. For TheRouter users this matters because qwen/text-embedding-v4 is the practical successor for new indexes, while v3 remains the compatibility model for existing vector spaces.
Use v3 when you already have vectors generated with v3 and cannot reindex immediately. Use v4 for new projects or migrations where you can re-embed all documents and queries with one model generation.
Yes. Alibaba Cloud documents dimensions support for text-embedding-v3 with 64, 128, 256, 512, 768, and 1,024 dimensions. Keep the same dimension across an index; changing it requires a new vector field or a reindex.
Yes. Alibaba documents an OpenAI-compatible embeddings API for text embedding models, and TheRouter exposes the same integration style through https://api.therouter.ai/v1 with model ID qwen/text-embedding-v3.
No. Treat each embedding model generation and dimension as a separate vector space. If you migrate from v3 to v4, rebuild the index with v4 embeddings rather than appending v4 vectors beside v3 vectors.
| source | URL | retrieved | |
|---|---|---|---|
| Model family | help.aliyun.com ↗ | 2026-07-24 | verified |
| Embedding dimensions | help.aliyun.com ↗ | 2026-07-24 | verified |
| Supported languages | help.aliyun.com ↗ | 2026-07-24 | verified |
| TheRouter context length | therouter.ai ↗ | 2026-07-24 | verified |
| TheRouter input price | therouter.ai ↗ | 2026-07-24 | verified |
| Alibaba Cloud public-cloud API compatibility | help.aliyun.com ↗ | 2026-07-24 | verified |
| Primary endpoint on TheRouter | therouter.ai ↗ | 2026-07-24 | verified |
| Successor model | qwenlm.github.io ↗ | 2026-07-24 | verified |
| Parameter count | — | — | unknown |
| Training cutoff | — | — | unknown |
| License | — | — | unknown |
| Alibaba Cloud documented embedding tasks | help.aliyun.com ↗ | 2026-07-24 | unknown |
| Qwen3 Embedding arrives as the v4 successor path | qwenlm.github.io ↗ | 2026-07-24 | verified |
| Should I use text-embedding-v3 or text-embedding-v4? | qwenlm.github.io ↗ | 2026-07-24 | to verify |
| Can I change the embedding dimension? | help.aliyun.com ↗ | 2026-07-24 | to verify |
| Does text-embedding-v3 support OpenAI-compatible clients? | help.aliyun.com ↗ | 2026-07-24 | to verify |
| Can I mix v3 and v4 vectors in the same vector database index? | qwenlm.github.io ↗ | 2026-07-24 | to verify |