One API for every model
The endpoint speaks the OpenAI format, so existing clients work with a new base URL. Switch models by changing one field.
See model pricesInference
Run models in production through one API.
Call open and frontier models such as DeepSeek V3.2 and Llama 4 Scout through one OpenAI-compatible API, or serve your own weights with vLLM on an H100.
What you can build
Three jobs people run here. Each on a product you can start today.
Self-host with the Inference crate
The Inference crate is a ready image your instance boots, with vLLM or SGLang one command away. Serve your weights on a GPU you pick.
Compare serving GPUsWeights that stay put
Keep model weights on a volume, so the next instance mounts them instead of downloading them again.
Read how volumes workExamples
Ask your agent. Or read the code it writes.
Connected by the line, Claude Code or Codex calls the Runcrate tools for you: launch, copy files, run, delete.
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
Works inClaude CodeCodexOpenCodeClineRoo CodeKilo CodeCursor
Models and tools
Models and engines people serve. Call them, or run them yourself.
| Name | Role | Runs on |
|---|---|---|
| DeepSeek V3.2 | Chat, code and reasoning | Inference API |
| Llama 4 Scout | Long-context open model | Instance or API |
| gpt-oss-120b | Open-weight reasoning with tools | Instance or API |
| Qwen3 235B Instruct | Multilingual, long context | Instance or API |
| vLLM | Serving engine for your weights | Instance |
| SGLang | Serving engine for your weights | Instance |
How it works
From first call to production. Three steps.
- 01
Pick your approach
Call a model through the Inference API, or launch an instance with the Inference crate when you need your own weights and settings.
- 02
Point your client at Runcrate
Set the base URL to https://api.runcrate.ai/v1 and use your API key. The OpenAI SDKs and the Runcrate SDKs both work.
- 03
Watch usage
See requests, tokens and spend in the usage dashboard. The API bills per token; instances bill for the time they run.
Pricing
What it costs. Live prices, as billed.
GPUs are billed for the time they run: by the second on most GPUs, by the minute on a few.
- A multi-node cluster for a month or longer: reserve one, and see every offer side by side.
Name
As billed
H100
GPU, 80GB, cheapest offer
$3.58/hr
L40S
GPU, 48GB, cheapest offer
$0.55/hr
DeepSeek V3.2
DeepSeek
$0.273 in, $0.399 out per 1M tokens
Llama 4 Scout 17B 16E
Meta
$0.105 in, $0.315 out per 1M tokens
gpt-oss 120B
OpenAI
$0.0389 in, $0.179 out per 1M tokens