Embedding models · 17 models · 6 makers
Embedding models: price, dimensions and max input
All AI modelsChecked 1 Nov 2026
17 embedding models from 6 makers in our catalogue. The cheapest API is text-embedding-3-small at $0.02 per 1M input tokens. Qwen3 Embedding 8B and Qwen3 VL Embedding 8B return the largest vectors (4,096 dimensions). Embed 5 Fast and Embed 5 Pro take the longest input (128,000 tokens). 8 models have open weights.
Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
Cite
Cite this page
ModelsAtlas. “Embedding models”. Retrieved 1 Nov 2026. https://modelsatlas.com/models/embeddings
- Models
- 17
- Cheapest / 1M
$0.02
derived We computed- Source
- Our catalogue
- Retrieved
- 1 Nov 2026
- Note
- Lowest standard-tier input price among embedding models in our catalogue.
- Most dimensions
- 4,096
- Longest input
- 128,000
9 with an API price
text-embedding-3-small
Qwen3 Embedding 8B and Qwen3 VL Embedding 8B
Embed 5 Fast and Embed 5 Pro · tokens
Embedding model APIs, cheapest first
USD per 1M input tokens
| # | Model | Price / 1M | Batch / 1M | Dimensions | Max input | Inputs |
|---|---|---|---|---|---|---|
| 1 | text-embedding-3-smallOpenAI | $0.02asserted Vendor states
| — | 1,536asserted Vendor states
| 8,192asserted Vendor states
| text |
| 2 | Embed 5 FastCohere | $0.08 | — | 2,048 | 128,000 | text and image |
| 3 | Mistral EmbedMistral AI | $0.10 | — | 1,024 | 8,192 | text |
| 4 | text-embedding-ada-002OpenAI | $0.10asserted Vendor states
| — | 1,536asserted Vendor states
| 8,192asserted Vendor states
| text |
| 5 | Embed 5 ProCohere | $0.12 | — | 2,048 | 128,000 | text and image |
| 6 | text-embedding-3-largeOpenAI | $0.13asserted Vendor states
| — | 3,072asserted Vendor states
| 8,192asserted Vendor states
| text |
| 7 | Codestral EmbedMistral AI | $0.15 | — | — | 8,192 | text |
| 8 | Gemini EmbeddingGoogle | $0.15 | $0.075 | 3,072asserted Vendor states
| 2,048asserted Vendor states
| text |
| 9 | Gemini Embedding 2Google | $0.20 | $0.10 | 3,072asserted Vendor states
| 8,192asserted Vendor states
| text, image, audio and video |
Open-source embedding models
Weights you can download and run yourself
| Model | Dimensions | Max input | Inputs | Licence |
|---|---|---|---|---|
| Qwen3 Embedding 8BQwen (Alibaba) · open weights | 4,096asserted Vendor states
| 32,768asserted Vendor states
| text | apache-2.0 |
| Qwen3 VL Embedding 8BQwen (Alibaba) · open weights | 4,096asserted Vendor states
| 32,768asserted Vendor states
| text, image and video | apache-2.0 |
| Qwen3 Embedding 4BQwen (Alibaba) · open weights | 2,560asserted Vendor states
| 32,768asserted Vendor states
| text | apache-2.0 |
| Qwen3 VL Embedding 2BQwen (Alibaba) · open weights | 2,048asserted Vendor states
| 32,768asserted Vendor states
| text, image and video | apache-2.0 |
| BGE Large EN v1.5BAAI · open weights | 1,024asserted Vendor states
| 512asserted Vendor states
| text | mit |
| BGE-M3BAAI · open weights | 1,024 | 8,192 | text | mit |
| Qwen3 Embedding 0.6BQwen (Alibaba) · open weights | 1,024asserted Vendor states
| 32,768asserted Vendor states
| text | apache-2.0 |
| EmbeddingGemma 300MGoogle · open weights | 768asserted Vendor states
| 2,048asserted Vendor states
| text | gemma |
Questions about embedding models
What is an embedding model?
A model that turns text (and for some models images, audio or video) into a vector of numbers, so that inputs with similar meaning get similar vectors. Embeddings power semantic search, retrieval for RAG, clustering and recommendations. You pay for input tokens only: there is no generated output.
What is the cheapest embedding model?
text-embedding-3-small at $0.02 per 1M input tokens on the standard tier is the cheapest embedding API we list. 8 open-weight models cost nothing per token if you run them yourself.
Which OpenAI embedding model should I use?
OpenAI lists text-embedding-3-small at $0.02 per 1M tokens (1,536 dimensions), text-embedding-ada-002 at $0.10 per 1M tokens (1,536 dimensions) and text-embedding-3-large at $0.13 per 1M tokens (3,072 dimensions). The cheaper model returns smaller vectors; the larger vectors cost more to store and search.
Which embedding models are open source?
Qwen3 Embedding 8B, Qwen3 VL Embedding 8B, Qwen3 Embedding 4B, Qwen3 VL Embedding 2B, BGE Large EN v1.5, BGE-M3, Qwen3 Embedding 0.6B and EmbeddingGemma 300M publish their weights. Check each licence on the model page before commercial use.
Which embedding model has the most dimensions and the longest input?
Qwen3 Embedding 8B and Qwen3 VL Embedding 8B return 4,096-dimension vectors by default. Embed 5 Fast and Embed 5 Pro accept up to 128,000 tokens per input. Many models can shorten their vectors on request; the table shows the default size.
What is the best embedding model?
We do not benchmark retrieval quality, so this page does not rank by it. Compare price, vector size, max input and input types here, then test the shortlist on your own data.
Keep going in your AI
Opens with this page as the source. We add nothing else to the prompt.
- What is an embedding model?
- What is the cheapest embedding model?
- Which OpenAI embedding model should I use?