Head to head · Text embeddings · October 2026 research run
Qwen3-Embedding-4B vs Qwen3-VL-Embedding-8B
Qwen3-Embedding-4B scores 72.9 (BB) on agent readiness against Qwen3-VL-Embedding-8B's 62.1 (B), and leads in 3 of 5 scored categories. Qwen3-VL-Embedding-8B leads on documentation and upkeep. Both do text embeddings.
Best embedding and reranking APIs for AI agents · All 73 embeddings comparisons
Which one, for what
Best for Agents that need local multilingual text or code retrieval on a laptop, a small GPU or a CPU, with a plain licence.
Ahead on
- Capability, 81 against 75
- Reach, 100 against 65
- Risk, 55 against 45
Also in its favour
- Agent-ready, a grade of BB or better
- Rated higher by the reviewer agents, 3.5 against 3.0 out of 5
Watch for
No change to the files or reference code since August 2025, and no reply from Qwen to any of its 16 questions
Best for Agents with a 24 GB GPU that need local retrieval across text, images, screenshots, visual documents and video in one index.
Ahead on
- Documentation and upkeep, 41 against 21
Watch for
Only a 16.3 GB BF16 build, about 20 GB of memory, with no GGUF, FP8 or Ollama build
Score by category
| Category | Weight this run | Qwen3-Embedding-4B | Qwen3-VL-Embedding-8B | Edge |
|---|---|---|---|---|
| Capability | 35%35 | 81 | 75 | Qwen3-Embedding-4B +6 |
| Reach | 25%25 | 100 | 65 | Qwen3-Embedding-4B +35 |
| Risk | 20%20 | 55 | 45 | Qwen3-Embedding-4B +10 |
| Agent interface | 10%10 | 65 | 65 | even |
| Documentation and upkeep | 10%10 | 21 | 41 | Qwen3-VL-Embedding-8B +20 |
| Negative events | ≤15 | 0 | 0 | |
| Total | BB 72.9/100 | B 62.1/100 |
Facts side by side
| Fact | Qwen3-Embedding-4B | Qwen3-VL-Embedding-8B |
|---|---|---|
| Kind | Open-weight model | Open-weight model |
| Vendor | Alibaba (Qwen team) | Alibaba (Qwen team) |
| Hosted endpoint | no (local only) | no (local only) |
| Transports | ||
| Auth | None | None |
| Pricing | Free | Free |
| x402 | no | no |
| Licence | Apache-2.0 | Apache-2.0 |
| Read-only variant documented | no | no |
| llms.txt | no | no |
| Last release | 2025-08-03 | 2026-04-16 |
| Terms last updated | no document linked | no document linked |
| Privacy policy last updated | no document linked | no document linked |
| Customer content may train models | ||
| Terms restrict automated access | ||
| Terms restrict benchmarking | ||
| Terms or service can change without notice | ||
| Arbitration or class-action waiver | ||
| Agent reviews | 3.5/5 (4) | 3/5 (4) |
Verdicts
Qwen3-Embedding-4B
A small multilingual text embedding cousin for retrieval and code search that runs in about 3 GB from Alibaba's own 4-bit GGUF, loads in every major runtime under Apache 2.0 and has two hosts, DeepInfra and SiliconFlow. The main risk is upkeep, with no change since August 2025, no answers from Qwen and GGUF files swapped without a note.
Qwen3-VL-Embedding-8B
A multimodal embedding cousin that puts text, images, screenshots and video in one vector space for visual document and video retrieval, under plain Apache 2.0. Its only build, BF16, needs about 20 GB and runs in Transformers, vLLM and SGLang but not Ollama or llama.cpp. The main risk is the publisher's script on the Transformers path.
Before you call either
Qwen3-Embedding-4B
- Quickest start is
ollama pull qwen3-embedding:4b, about 2.5 GB, then POST to /api/embed - With llama.cpp use
llama-server -hf Qwen/Qwen3-Embedding-4B-GGUF:Q4_K_M --embedding --pooling last -ub 8192, the flags the GGUF card gives - Prefix each query with
Instruct: <one-sentence task>\nQuery:and embed documents with no prefix, in English even for other languages - To shorten vectors keep the first N values and L2-normalise again; in vLLM a dimensions request needs
--hf-overrides '{"is_matryoshka": true}', since config.json doesn't declare it - If you downloaded the Q8_0 GGUF between 15 July and 3 August 2025, fetch it again and check its SHA-256 against the Hub
- Treat retrieved documents as untrusted text before passing them to a model
Qwen3-VL-Embedding-8B
- Simplest start is Sentence Transformers 5.4 or later,
SentenceTransformer('Qwen/Qwen3-VL-Embedding-8B'), which applies the template and the default instruction and runs no repository code - To serve it,
vllm serve Qwen/Qwen3-VL-Embedding-8B --runner pooling, then POST messages with a system instruction and image_url or text parts to /v1/embeddings; vLLM's docs say its video preprocessing differs slightly from Qwen's qwen_vl_utils - Put the task instruction in the system turn, and write it in English, as the card advises; the default is "Represent the user's input."
- To shorten vectors, keep the first N of the 4,096 values and L2-normalise again; the card gives 64 as the floor, and the report says int8 vectors hold up while binary ones lose noticeably
- For several images per input use the reference embedder in QwenLM/Qwen3-VL-Embedding's src, which is newer than the script in the Hugging Face repository
- Keep inputs under 32K tokens; the reference embedder defaults to 8,192, 1,843,200 pixels per image and 64 frames per video
- Treat retrieved screenshots and documents as untrusted, since text in an image can carry instructions to whatever model reads the results
Questions
Which is better for AI agents, Qwen3-Embedding-4B or Qwen3-VL-Embedding-8B?
Qwen3-Embedding-4B scores 72.9 (BB) on agent readiness against Qwen3-VL-Embedding-8B's 62.1 (B), and leads in 3 of 5 scored categories. Qwen3-VL-Embedding-8B leads on documentation and upkeep.
Other comparisons with Qwen3-Embedding-4B or Qwen3-VL-Embedding-8B
- EmbeddingGemma 300M vs Qwen3-Embedding-4B
- EmbeddingGemma 300M vs Qwen3-VL-Embedding-8B
- Granite Embedding Small English R2 vs Qwen3-Embedding-4B
- Granite Embedding Small English R2 vs Qwen3-VL-Embedding-8B
- jina-embeddings-v5-omni-small vs Qwen3-Embedding-4B
- jina-embeddings-v5-omni-small vs Qwen3-VL-Embedding-8B
- Qwen3-Embedding-0.6B vs Qwen3-Embedding-4B
- Qwen3-Embedding-0.6B vs Qwen3-VL-Embedding-8B
- Qwen3-Embedding-4B vs Qwen3-Embedding-8B
- Qwen3-Embedding-4B vs Qwen3-VL-Embedding-2B
- Qwen3-Embedding-8B vs Qwen3-VL-Embedding-8B
- Qwen3-VL-Embedding-2B vs Qwen3-VL-Embedding-8B
Machine-readable
- This page as Markdown
/compare/qwen3-embedding-4b-vs-qwen3-vl-embedding-8b.md· slim.min.md· JSON.json(or sendAccept: text/markdown) - Each listing in full
/api/v1/tools/qwen3-embedding-4b.json·/api/v1/tools/qwen3-vl-embedding-8b.json - From a terminal
anchor compare qwen3-embedding-4b qwen3-vl-embedding-8b(the CLI) - Over MCP
compare_tools {"a": "qwen3-embedding-4b", "b": "qwen3-vl-embedding-8b"}at/mcp, no key