Open source LLM API
for production workloads
One OpenAI-compatible API to open-weight models. Your token price is locked when you top up, and a real team answers when you scale.
Served models
A needs-driven catalog of open-weight models — we serve what teams actually run, and add models by demand.
GLM-5.3-Flash
by Z.ai · open weights
zai-org/GLM-5.3-Flash - Context
- 400K tokens
- Max output
- 16K tokens
- Input / 1M
- $0.20
- Output / 1M
- $0.20
Agentic coding and multi-step workflows, the efficient tier of the GLM line. Tool calling supported, served in fp8.
DeepSeek V4 Flash
by DeepSeek · open weights
deepseek-ai/DeepSeek-V4-Flash-0731 - Context
- 400K tokens
- Max output
- 16K tokens
- Input / 1M
- $0.20
- Output / 1M
- $0.20
Fast answers by default, switchable reasoning per request for harder tasks. Tool calling supported, served in fp8.
MiniMax M2
by MiniMax · open weights
MiniMaxAI/MiniMax-M2.7 - Context
- 205K tokens
- Max output
- 16K tokens
- Input / 1M
- $0.25
- Output / 1M
- $0.25
Efficient coding and agent workloads, with built-in step-by-step thinking. Tool calling supported, served in fp8.
Also served: BGE-M3 embeddings (BAAI/bge-m3) for RAG and search — see the Embeddings & RAG guide.
Full model referenceYour price is locked when you top up
Usage-based APIs surprise you at the end of the month. TrustedRails works the other way: you top up in USD, the per-token rate is fixed for that balance, and your budget stays exactly where you set it.
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| GLM-5.3-Flash | $0.20 | $0.20 |
| DeepSeek V4 Flash | $0.20 | $0.20 |
| MiniMax M2 | $0.25 | $0.25 |
List prices as of September 2026. List prices may change over time — the rate you topped up at stays yours.
Estimate your monthly bill in the cost calculatorNo vendor lock-in
Easy to adopt
The API is fully OpenAI-compatible: change the
base_url and the API key in
the SDK, tool, or framework you already use, and your existing code
keeps working.
Just as easy to leave
The same two lines take you to any other vendor — and the models are open-weight, so you can even run them yourself. We keep customers with price and service, not contracts.
What teams run on open models
Coding & agents
Agentic coding in Cline, Cursor, or Zed on an open model with tool calling and a 400K context.
Chat & assistants
Production chatbots and internal assistants through the same API your OpenAI code already speaks.
RAG & search
Retrieval pipelines end to end — chat models plus served BGE-M3 embeddings, one endpoint.
Automation
LLM steps in n8n, background jobs, batch processing — predictable per-token cost at volume.
Works in your tools
Anything with an OpenAI-compatible endpoint setting connects to TrustedRails. Step-by-step guides, each verified against the live tool:
…and any other OpenAI SDK or client. Can't find your tool? Tell us — we write the guide.
Connect in two lines
Change the base URL and API key in any OpenAI client — that's the whole integration.
Python
Official openai package —
works with LangChain, LlamaIndex, and any OpenAI-compatible framework.
from openai import OpenAI
client = OpenAI(
base_url="https://proxy.trustedrails.com/v1",
api_key="tr-prx-YOUR_KEY_HERE"
)
response = client.chat.completions.create(
model="zai-org/GLM-5.3-Flash",
messages=[{"role": "user", "content": "Hello!"}]
)JavaScript / TypeScript
Official openai npm package —
Node.js, Deno, Bun, and browser environments.
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://proxy.trustedrails.com/v1",
apiKey: "tr-prx-YOUR_KEY_HERE",
});
const response = await client.chat.completions.create({
model: "zai-org/GLM-5.3-Flash",
messages: [{ role: "user", content: "Hello!" }],
});Talk to people, not a rate limiter
Running millions of tokens a day? We work sales-led: volume terms, capacity planning, and a direct line to the team that operates the infrastructure — not a ticket queue.
- Volume terms with your rate locked at top-up
- Models added to the catalog by your need
- No lock-in — leave as easily as you arrived
Your data never reaches model vendors
Your prompts and completions are not used for training. Inference runs on distributed independent operators — no model vendor sees your data.
FAQ
What is TrustedRails?
TrustedRails is an OpenAI-compatible API for open-source LLMs. You point any OpenAI SDK or tool at our endpoint, pick an open model, and pay a USD token price that is locked in when you top up your balance.
Which models can I use?
Currently served: GLM-5.3-Flash (zai-org/GLM-5.3-Flash), DeepSeek V4 Flash (deepseek-ai/DeepSeek-V4-Flash-0731), and MiniMax M2 (MiniMaxAI/MiniMax-M2.7) for chat and agents, plus BGE-M3 for embeddings. The catalog is driven by demand: we add open models our customers need.
Is it OpenAI-compatible?
Yes. The API implements the OpenAI chat-completions format, so official OpenAI SDKs, LangChain, LlamaIndex, and tools with a custom-endpoint setting work by changing the base URL and API key.
How does locked pricing work?
When you top up, the current per-token USD rate is fixed for the balance you bought. List prices can change over time, but tokens you already paid for never get more expensive — no surprise usage bills.
Am I locked in?
No. Because the API is OpenAI-compatible and the models are open-weight, moving to TrustedRails is a two-line change — and moving away is just as easy. We keep customers with price and service, not lock-in.
Is my data used for training?
No. Your prompts and completions are not used for training. Inference runs on distributed independent operators, so no model vendor sees your data.
Run open models in production
OpenAI-compatible API, price locked at top-up, no lock-in.
Get an API keyAlready have an account? Sign in