70% off Wan3.0 Standard and 50% off Wan3.0 Prime. Try Wan3.0
The OpenRouter alternative with guaranteed capacity.
LLM inference without the platform fee. Every major lab's models answer on one OpenAI-compatible API, with dependable capacity as you scale.
- 52LLMs on one API key
- 10model labs
- 0%platform fee
- 1bill for LLMs, image, video and audio
OpenRouter and Runware, side by side
Gateways pass each request to an inference provider. Runware is one of those providers, on Vercel AI Gateway and Requesty among others, and you can call it directly.
| OpenRouter | Runware | |
|---|---|---|
| Platform fee | 5.5% | None |
| Who serves the model | A third-party provider | Runware, or the model's own lab |
| Rate limits | Set by each provider | Raised to match your reserved capacity |
| Your own models | An Enterprise add-on | Serverless GPUs or dedicated capacity |
| Image, video and audio | 140+ models | 400K+ models at the lowest cost |
| Model routing | Auto Router, chat only | Model Router, text and media |
Every major LLM on one API key
Claude, GPT, Gemini, Kimi, DeepSeek and GLM, with 400K+ image, video and audio models on the same key and the same invoice.
Dependable capacity as you scale
Reserve dedicated LLM inference for your workload, with throughput, concurrency and latency commitments.
Request dedicated capacity- Guaranteed throughput
- Capacity reserved for your traffic, with rate limits raised to match.
- Custom SLAs
- Volume pricing and service levels set for your deployment.
- Your models
- Open-weight models, or your own custom and fine-tuned ones.
- Region lock
- Keep inference in the US or in the EU, with zero data retention.
Run your own models and pipelines on Serverless GPUs
Deploy an open or fine-tuned LLM from a template on its own OpenAI-compatible endpoint, or bring a container for image, video and multimodal pipelines. Both scale to zero when idle.
Or set "model": "auto" and let the Model Router pick
Runware's Model Router chooses a model for each request by quality, speed and cost. Routing is free.
Your request
{ "taskType": "textInference", "model": "auto", "includeCost": true, "routing": { "preferences": { "quality": 3, "speed": 4, "cost": 5 }, "maxPrice": 0.01, "requirements": "image input" } }The router
- Claude Opus 4.8Over your price ceiling
- GLM-5.1Does not meet the requirement
- GPT-5.4 MiniRanked second
- Gemini 3 FlashPicked
The response
Gemini 3 Flash
{ "taskType": "textInference", "model": "google:gemini@3-flash", "cost": 0.0015 }
Priorities
Weight quality, speed and cost from 0 to 5. The weights order the models that fit.
A price ceiling
Set maxPrice in USD. Models above it are ruled out before ranking.
Written requirements
Say what no parameter can hold, such as "an open-source model", in up to 3,000 characters.
Allow and block lists
Keep the router to models you approve, or exclude the ones you do not.
Switching from OpenRouter: three values change
Runware's LLM API accepts the OpenAI Chat Completions format. Keep the OpenAI SDK and change the base URL, the key and the model ID.
from openai import OpenAI
client = OpenAI(
api_key="your_runware_api_key",
base_url="https://api.runware.ai/v1",
)
response = client.chat.completions.create(
model="deepseek:v4.1@flash",
messages=[{"role": "user", "content": "Hi"}],
stream=True,
)Frequently asked questions
Is Runware an OpenRouter alternative?
Yes. Runware is an alternative to OpenRouter for LLM inference, with one key for many labs' models. It also sells dedicated capacity and runs your own models.
Does OpenRouter charge a fee?
Yes. OpenRouter adds a 5.5% platform fee when you buy credits. Runware charges none.
Is the Runware API OpenAI-compatible?
Yes. The API accepts the OpenAI Chat Completions format, so you keep the OpenAI SDK and change the base URL, the key and the model ID.
How do I get dedicated capacity?
Open the form with your models, volume and region. Runware sizes the capacity and replies with pricing and service levels.
Leave the gateway. Keep your code.
Get an API key and make your first call today, or ask for dedicated capacity sized to your workload.