How free embeddings work in FreeLLMAPI
The /v1/embeddings endpoint uses the OpenAI request format. Any OpenAI SDK can call it. You send text, and you get one vector for each input string.
Embeddings use a different failover rule from chat. Chat requests can move to a different model. Embedding requests never move to a different model. Vectors from two models are not compatible, so a silent model change would corrupt your vector store.
- Family: FreeLLMAPI groups providers by family. A family is one model identity with one vector size, for example
bge-m3at 1024 dimensions. - Failover: when one provider in a family hits its limit, the router tries the next provider that serves the same family.
- Default:
model: "auto"uses the default family. You set the default family on the Models, Embeddings page of the dashboard.
Free embedding models today
The table lists every free embedding model in the FreeLLMAPI catalog. The data comes from the live catalog. "Max input" is the maximum number of tokens for one input string. "Free limit" is the free quota that the provider publishes.
| Model | Provider | Dimensions | Max input | Free limit |
|---|---|---|---|---|
Gemini Embeddinggemini-embedding-001 | Google AI Studio | 3072 | 2048 | 100 rpm · 1K req/day |
Nemotron Embed VL 1Bnvidia/llama-nemotron-embed-vl-1b-v2 | NVIDIA NIM | 2048 | 8192 | ~40 rpm |
Nemotron 3 Embed 1B (OR free)nvidia/nemotron-3-embed-1b:free | OpenRouter | 2048 | 32768 | 20 rpm · shared free daily cap |
SEA-LION E5 Embedding 600Maisingapore/SEA-LION-E5-Embedding-600M | SEA-LION | 1024 | 8192 | free · 10 rpm |
SEA-LION ModernBERT Embedding 300Maisingapore/SEA-LION-ModernBERT-Embedding-300M | SEA-LION | 1024 | 8192 | free · 10 rpm |
SEA-LION ModernBERT Embedding 600Maisingapore/SEA-LION-ModernBERT-Embedding-600M | SEA-LION | 1024 | 8192 | free · 10 rpm |
BGE-M3@cf/baai/bge-m3 | Cloudflare Workers AI | 1024 | 8192 | 10K neurons/day (shared) |
EmbeddingGemma 300M@cf/google/embeddinggemma-300m | Cloudflare Workers AI | 768 | 2048 | 10K neurons/day (shared) |
Qwen3 Embedding 0.6B@cf/qwen/qwen3-embedding-0.6b | Cloudflare Workers AI | 1024 | 8192 | 10K neurons/day (shared) |
Cohere Embed v4embed-v4.0 | Cohere | 1536 | 128000 | 1K calls/mo (shared w/ chat) |
Gemini Embedding 2gemini-embedding-2 | Google AI Studio | 3072 | 8192 | free-tier rate limited |
Nemotron 3 Embed 1Bnvidia/nemotron-3-embed-1b | NVIDIA NIM | 2048 | 4096 | ~40 rpm |
PLaMo Embedding 1B (CF)@cf/pfnet/plamo-embedding-1b | Cloudflare Workers AI | 2048 | 4096 | 10K neurons/day (shared) |
BGE Small EN v1.5 (CF)@cf/baai/bge-small-en-v1.5 | Cloudflare Workers AI | 384 | 512 | 10K neurons/day (shared) |
BGE Base EN v1.5 (CF)@cf/baai/bge-base-en-v1.5 | Cloudflare Workers AI | 768 | 512 | 10K neurons/day (shared) |
BGE Large EN v1.5 (CF)@cf/baai/bge-large-en-v1.5 | Cloudflare Workers AI | 1024 | 512 | 10K neurons/day (shared) |
Mistral Embedmistral-embed | Mistral | 1024 | 8192 | free mode · monthly cap (shared) |
Codestral Embedcodestral-embed | Mistral | 1536 | 8192 | free mode · monthly cap (shared) |
Llama Text Embed v2 (Pinecone)llama-text-embed-v2 | Pinecone | 1024 | 2048 | 5M tokens/mo per model |
Multilingual E5 Large (Pinecone)multilingual-e5-large | Pinecone | 1024 | 507 | 5M tokens/mo per model |
Nemotron 3 Embed 1B (Speka)nvidia/nemotron-3-embed-1b | Speka | 2048 | 8192 | $1/month shared across Speka models |
OpenAI: Text Embedding 3 Small (ElectronHub)text-embedding-3-small | ElectronHub | 1536 | 8192 | $0.25/week shared free credits |
OpenAI: Text Embedding 3 Large (ElectronHub)text-embedding-3-large | ElectronHub | 3072 | 8192 | $0.25/week shared free credits |
Nemotron 3 Embed 8B (Inferbase)nvidia-nemotron-3-embed-8b | Inferbase | 4096 | 8192 | 5,000 requests/month; 100 requests and $0.50 usage/day shared |
Qwen 3 Embedding 0.6B (Inferbase)qwen3-embedding-0-6b | Inferbase | 1024 | 8192 | 5,000 requests/month; 100 requests and $0.50 usage/day shared |
Qwen 3 Embedding 4B (Inferbase)qwen3-embedding-4b | Inferbase | 2560 | 8192 | 5,000 requests/month; 100 requests and $0.50 usage/day shared |
Qwen 3 Embedding 8B (Inferbase)qwen3-embedding-8b | Inferbase | 4096 | 8192 | 5,000 requests/month; 100 requests and $0.50 usage/day shared |
LFM2.5 Embedding 350M (LLMTR)liquid/lfm-2.5-embedding-350m-free | LLMTR | 1024 | 512 | Ongoing zero-price route; provider terms apply |
BGE-M3 (SEA-LION)BAAI/bge-m3 | SEA-LION | 1024 | 8192 | free · 10 rpm |
Cohere Embed English v3embed-english-v3.0 | Cohere | 1024 | 512 | 1K calls/mo (shared w/ chat) |
Qwen3 Embedding 8B (OVH)Qwen3-Embedding-8B | OVH AI Endpoints | 4096 | 32768 | free · anonymous 2 req/min per IP |
BGE-M3 (OVH)bge-m3 | OVH AI Endpoints | 1024 | 8192 | free · anonymous 2 req/min per IP |
BGE Multilingual Gemma2 (OVH)bge-multilingual-gemma2 | OVH AI Endpoints | 3584 | 8192 | free · anonymous 2 req/min per IP |
Mistral: Codestral Embed (ElectronHub)codestral-embed | ElectronHub | 1536 | 8192 | $0.25/week shared free credits |
Mistral: Mistral Embed (ElectronHub)mistral-embed | ElectronHub | 1024 | 8192 | $0.25/week shared free credits |
Google: Gemini Embedding 2 (ElectronHub)gemini-embedding-2 | ElectronHub | 3072 | 8192 | $0.25/week shared free credits |
Google: Gemini Embedding 001 (ElectronHub)gemini-embedding-001 | ElectronHub | 3072 | 2048 | $0.25/week shared free credits |
OpenAI: Text Embedding Ada 002 (ElectronHub)text-embedding-ada-002 | ElectronHub | 1536 | 8192 | $0.25/week shared free credits |
Cohere Embed English Light v3embed-english-light-v3.0 | Cohere | 384 | 512 | 1K calls/mo (shared w/ chat) |
Cohere Embed Multilingual v3embed-multilingual-v3.0 | Cohere | 1024 | 512 | 1K calls/mo (shared w/ chat) |
Cohere Embed Multilingual Light v3embed-multilingual-light-v3.0 | Cohere | 384 | 512 | 1K calls/mo (shared w/ chat) |
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Some providers share one quota between chat and embeddings. For example, a Cohere trial key allows 1,000 API calls each month for all endpoints together. Cloudflare Workers AI gives 10,000 neurons each day, shared by all Workers AI models on the account.
Step 1: Install FreeLLMAPI and add keys
FreeLLMAPI is open source. It runs on your computer in Docker. Your provider keys stay on your computer.
Install (macOS or Linux)
curl -fsSL https://freellmapi.co/install.sh | bashOpen the dashboard at http://localhost:3001. Add your provider keys on the Keys page. For embeddings, a Google AI Studio key and a Cloudflare key give a good start. Copy your unified key from the same page.
Step 2: Call the embeddings endpoint
Send a list of strings to get several vectors in one request. Use the same model value for your documents and for your search queries.
curl
curl http://localhost:3001/v1/embeddings \
-H "Authorization: Bearer <unified-key>" \
-H "Content-Type: application/json" \
-d '{"model": "bge-m3", "input": ["the quick brown fox", "a fast brown fox"]}'Python: embed and compare two texts
import math
from openai import OpenAI
client = OpenAI(base_url="http://localhost:3001/v1", api_key="<unified-key>")
resp = client.embeddings.create(
model="bge-m3", # a family name, "auto", or a provider model id
input=["How do I reset my password?", "Steps to change a forgotten password"],
)
a, b = resp.data[0].embedding, resp.data[1].embedding
print(len(resp.data), "vectors of", len(a), "dimensions")
dot = sum(x * y for x, y in zip(a, b))
norm = math.sqrt(sum(x * x for x in a)) * math.sqrt(sum(y * y for y in b))
print("cosine similarity:", round(dot / norm, 3))The model field accepts three forms: auto, a family name such as bge-m3, or a provider model id such as @cf/baai/bge-m3. A provider model id resolves to its family, so failover still stays inside that family.
Which free embedding model to choose
Choose the model once for each vector store. If you change the model later, you must embed all documents again.
| Need | Model family | Why |
|---|---|---|
| Short English or multilingual text, many providers | bge-m3 | 1024 dimensions, 8,192 input tokens. Several free providers serve it, so failover has more options. |
| High quality retrieval with a Google key | gemini-embedding-001 | 3072 dimensions. The input limit is 2,048 tokens, so split long documents into chunks. |
| Long documents in one vector | embed-v4.0 (Cohere) | Large input window. The trial key allows only 1,000 calls each month, shared with chat. |
| Small vectors and fast search | bge-small-en-v1.5 | 384 dimensions. English only, and the input limit is 512 tokens. |
| Code search | codestral-embed | Trained for code. It uses the Mistral free mode quota. |
Large vectors cost more storage and memory in the vector database. A 3072-dimension vector is three times larger than a 1024-dimension vector. For most RAG projects under one million chunks, 768 to 1024 dimensions give good results.
Limits and pitfalls
- Input limits: text above the "Max input" value can be cut or rejected. Split documents into chunks of 300 to 800 tokens before you embed them.
- One family per store: do not mix vectors from two families in one index. The similarity scores have no meaning across models.
- Normalization: Google says that
gemini-embedding-001vectors with fewer than 3072 dimensions must be normalized by you. Normalize all vectors before you use dot product as the score. - Rate limits: free tiers count requests per minute or per day. Send batches of strings in one request, not one request for each string.
- Anonymous tiers: OVH AI Endpoints allows 2 requests per minute for each IP address and model without a key. Use it as a fallback, not for bulk indexing.
Troubleshooting
| Problem | Cause | Fix |
|---|---|---|
| Error 429 during bulk indexing | The provider limit for the family is reached | Add a second provider key for the same family, or send fewer requests each minute |
| Search results are wrong after a change | Documents and queries use different families | Use one model value for both. Embed the documents again if you changed it |
| Vector length is not what you expect | auto uses the default family | Set an explicit family name, or check the default on Models, Embeddings |
| Long texts give poor matches | The text is longer than the input limit | Split the text into smaller chunks |
| Error 401 | The unified key is wrong or missing | Copy the unified key again from the Keys page |
Frequently asked questions
Is there a free embeddings API?
Yes. Google AI Studio, Cloudflare Workers AI, Cohere, NVIDIA, Mistral, OVH and others give free embedding models with rate limits. FreeLLMAPI puts 41 of them behind one OpenAI-compatible endpoint.
Can I use the OpenAI SDK for free embeddings?
Yes. Set the base URL to http://localhost:3001/v1 and use your unified key. Then call client.embeddings.create with a family name such as bge-m3.
Why does FreeLLMAPI not fail over to a different embedding model?
Vectors from different models are not compatible. A silent model change would break similarity search. The router only fails over between providers that serve the same model family.
What is the best free embedding model for RAG?
For most projects, bge-m3 is a good default. It has 1024 dimensions, an 8,192-token input limit and several free providers. Use gemini-embedding-001 if you want 3072 dimensions and have a Google key.
How many dimensions should my embeddings have?
Use 768 to 1024 dimensions for most search and RAG work. Larger vectors use more storage and memory. Choose the size once, because a change means you must embed all documents again.
Related guides
- Free LLM API for RAG
- Free LLM API for LangChain
- Free LLM API in Python
- free cohere api
- Free LLM API rate limits
Sources
- FreeLLMAPI docs: Embeddings (checked 20 November 2026)
- Google AI Studio docs: Embeddings (checked 20 November 2026)
- Cloudflare Workers AI pricing (checked 20 November 2026)
- Cohere docs: API keys and rate limits (checked 20 November 2026)
- OVHcloud AI Endpoints capabilities (checked 20 November 2026)