Features / Endpoints
Embeddings, audio and batches, one key.
Embed and rerank your own documents for long-term memory, transcribe voice notes, and queue thousands of requests as one batch, all on the key that runs your uncensored chat models.
Embeddings /v1/embeddings
| Model | Model id | Developer | Price per million tokens |
|---|---|---|---|
| BGE-M3 | bge-m3 | BAAI | $0.0295 |
| Qwen3 Embedding 0.6B | qwen3-embedding-0.6b | Qwen | $0.0295 |
| BGE-EN-ICL | bge-en-icl | BAAI | $0.0313 |
| Multilingual E5 Large Instruct | multilingual-e5-large-instruct | intfloat | $0.0313 |
| Qwen3 Embedding 8B | qwen3-embedding-8b | Qwen | $0.0313 |
| BGE Small EN v1.5 | bge-small-en-v1.5 | BAAI | $0.0505 |
| Text Embedding Ada 002 | text-embedding-ada-002 | OpenAI | $0.12 |
| Text Embedding 3 Large | text-embedding-3-large | OpenAI | $0.13 |
| Gemini Embedding 001 | gemini-embedding-001 | $0.15 | |
| BGE Base EN v1.5 | bge-base-en-v1.5 | BAAI | $0.1665 |
| Text Embedding 3 Small | text-embedding-3-small | OpenAI | $0.4363 |
Rerank /v1/rerank
| Model | Model id | Developer | Price per million tokens |
|---|---|---|---|
| BGE Reranker Base | bge-reranker-base | BAAI | $0.0078 |
Speech to text /v1/audio/transcriptions
| Model | Model id | Developer | Price per audio minute |
|---|---|---|---|
| Whisper Large v3 Turbo | whisper-large-v3-turbo | OpenAI | $0.0013 |
| Parakeet TDT 0.6B v3 | parakeet-tdt-0.6b-v3 | NVIDIA | $0.015 |
| Whisper Large v3 | whisper-large-v3 | OpenAI | $0.015 |
Embeddings and speech to text with the OpenAI SDK
from openai import OpenAI client = OpenAI(base_url="https://api.uncensoredllm.ai/v1", api_key="sk-ullm-live-…") vectors = client.embeddings.create(model="bge-m3", input=["first passage", "second passage"])text = client.audio.transcriptions.create(model="whisper-large-v3-turbo", file=open("session.mp3", "rb"))print(len(vectors.data[0].embedding), text.text)Rerank
The documents scored against the query, best first; the query counts once beside every document.
curl https://api.uncensoredllm.ai/v1/rerank \ -H "Authorization: Bearer sk-ullm-live-…" \ -H "Content-Type: application/json" \ -d '{ "model": "bge-reranker-base", "query": "Which note mentions the lighthouse?", "documents": ["The keeper left at dawn.", "A light swept the cliffs every nine seconds.", "Supper was cold."], "top_n": 2 }'Batch API
One line per request, each with its own custom_id; up to 50 MB and 50,000 lines a file, kept 30 days.
{"custom_id": "ch-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "glm-5.3-uncensored", "messages": [{"role": "user", "content": "Write chapter 1."}]}}{"custom_id": "ch-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "glm-5.3-uncensored", "messages": [{"role": "user", "content": "Write chapter 2."}]}}upload = client.files.create(file=open("chapters.jsonl", "rb"), purpose="batch")batch = client.batches.create(input_file_id=upload.id, endpoint="/v1/chat/completions", completion_window="24h") batch = client.batches.retrieve(batch.id)if batch.status == "completed": print(client.files.content(batch.output_file_id).text)Image edits take OpenAI’s request at /v1/images/edits on the image models that list it. Every snippet is in the console docs.
Questions
BGE-M3, Qwen3 Embedding 0.6B, BGE-EN-ICL, Multilingual E5 Large Instruct, Qwen3 Embedding 8B, BGE Small EN v1.5, Text Embedding Ada 002, Text Embedding 3 Large, Gemini Embedding 001, BGE Base EN v1.5, Text Embedding 3 Small. Each takes OpenAI's embeddings request and is billed per input token at the price in the table.
Per minute of audio, rounded up to the second, at the model’s price. Upload mp3, mp4, m4a, wav, webm, ogg, flac or aac up to 25 MB; /v1/audio/translations turns speech into English text the same way.
Batch lines run at the normal catalog price, through the same routes as a direct request, within 24 hours. Use it to send thousands of requests without holding connections open: upload a JSONL file, create the batch, read the output file.
Yes. Every line names its model, so a batch can draft chapters, rewrite cards or label data on any listed chat model, or embed text, from one file of up to 50,000 lines.
No. The same sk-ullm- key and prepaid balance pay for chat, embeddings, rerank, speech to text, batches and image edits.