HUGGING FACE ALTERNATIVE
Hub model ids, without hosting them.
Keep your huggingface_hub or OpenAI client, point it at Runcrate, and call DeepSeek, Qwen, Llama, Gemma and Mistral by their Hub ids, billed per token.
Hand this to your agent
Add a Runcrate model to this project: run curl -fsSL"https://runcrate.ai/agents.md?intent=build"and follow it.
Your agent asks for an API key. Create a key
Quick start
Hand it to your agent. Or read the request it sends.
Your agent writes and runs it. The same key works for every model and every GPU.
Available models
Hub ids you can call today. Billed prices, as charged.
| Model | Maker | Price | Detail |
|---|---|---|---|
| Qwen/Qwen3-235B-A22B-Instruct-2507 | Alibaba | $0.0945 in, $0.578 out per 1M tokens | Same id as the Hub repo, multilingual |
| deepseek-ai/DeepSeek-V3.2 | DeepSeek | $0.273 in, $0.399 out per 1M tokens | Same id as the Hub repo, reasoning and code |
| meta-llama/Llama-4-Scout-17B-16E-Instruct | Meta | $0.105 in, $0.315 out per 1M tokens | Same id as the Hub repo, multimodal MoE |
| google/gemma-3-27b-it | $0.084 in, $0.168 out per 1M tokens | Same id as the Hub repo, instruction-tuned | |
| mistralai/Mistral-Small-3.2-24B-Instruct-2506 | Mistral AI | $0.0788 in, $0.21 out per 1M tokens | Same id as the Hub repo, image input |
| openai/whisper-large-v3 | OpenAI | $0.000473 per minute | Same id as the Hub repo, speech to text |
Why Runcrate
Built for production. One key, one balance.
Hub ids in the model field
Models such as Qwen/Qwen3-235B-A22B-Instruct-2507 and deepseek-ai/DeepSeek-V3.2 are called by their Hub repo id, so code that names Hub models keeps its strings.
Your client stays
huggingface_hub's InferenceClient takes a base_url: set it to https://api.runcrate.ai/v1 with a Runcrate key and keep calling client.chat.completions.create().
Nothing to deploy
The models are already serving. Send a request with the model id; there is no endpoint to create, scale or keep warm.
Any other Hub model on a GPU
For a model the catalog does not serve, start an H100 from $3.88/hr on the same balance and serve the Hub weights with vLLM.
Comparison
Runcrate and Hugging Face Inference Providers, side by side.
| Feature | Runcrate | Hugging Face Inference Providers |
|---|---|---|
| Model ids | Hub repo ids for the models above | Hub repo ids, routed to a partner provider |
| Price per request | The billed price in the table above | The provider's rate, passed through with no Hugging Face markup |
| Included credits | Five open models free with a GitHub account, then a prepaid balance | $2.00 a month on PRO; free accounts buy credits |
| GPUs on the same balance | Yes, from $3.88/hr for an H100 | Credits also cover Inference Endpoints and Spaces hardware |
FAQ
Common questions.
Related