Chat, reasoning and code
Support bots, coding assistants and reasoning tasks with DeepSeek V3.2, DeepSeek R1 or DeepSeek V4 Pro, streamed token by token.
See model pricesLanguage models
Language models behind one API.
Call DeepSeek V3.2, DeepSeek V4.1 Flash, DeepSeek V4 Pro and Llama 4 Scout with one OpenAI-compatible request. Switch by changing one field.
What you can build
Three jobs people run here. Each on a product you can start today.
Long documents in one prompt
DeepSeek V4.1 Flash reads up to 1M tokens at once: a codebase, a contract set or a book.
See model pricesHost an open model yourself
The Inference crate is a ready image your instance boots, with vLLM one command away. Serve Llama or Mistral on a GPU you pick.
Compare serving GPUsExamples
Ask your agent. Or read the code it writes.
Claude Code or Codex sends the request for you. Any OpenAI-compatible client works with one base URL.
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
Language models people pick. Switch with one parameter.
| Name | Role | Runs on |
|---|---|---|
| DeepSeek V4 Pro | Agents, code, analysis | Inference API |
| DeepSeek V3.2 | Code, math and reasoning | Inference API |
| DeepSeek V4.1 Flash | Fast, 1M-token context | Inference API |
| Llama 4 Scout | Open-weight, long context | Instance or API |
| Qwen3 235B Instruct | Multilingual, reasoning | Instance or API |
| Mistral Small 3.2 | Efficient everyday model | Instance or API |
How it works
From first call to production. Three steps.
- 01
Pick your model
DeepSeek, Llama, Qwen, gpt-oss, Mistral and more. Each suits a different job: reasoning, code, speed or cost.
- 02
Call one API
One endpoint, one format. Send prompts with streaming, function calling or structured outputs, and switch models with one parameter.
- 03
Ship and watch spend
Pay per token with no minimum. See requests, tokens and spend per model in the usage dashboard.
Pricing
What it costs. Live prices, as billed.
Models are billed per token, image or second of output, at the price shown.
- A multi-node cluster for a month or longer: reserve one, and see every offer side by side.
Name
As billed
DeepSeek V3.2
DeepSeek
$0.273 in, $0.399 out per 1M tokens
DeepSeek V4.1 Flash
DeepSeek
$0.147 in, $0.441 out per 1M tokens
Llama 4 Scout 17B 16E
Meta
$0.105 in, $0.315 out per 1M tokens