A reliable small model with a huge quota
Frontier models are expensive and rate limited. Most everyday work doesn't need them. Cabrio serves Qwen3.6-36B and DeepSeek V4 Flash behind an OpenAI compatible API: billions of tokens a month, a flat price, total privacy. Enough to run your agents around the clock.
Pre-registerNo payment now. We email you when the beta opens.
Most work doesn't need a frontier model
Frontier providers sell their best models with tight quotas at high prices. For hard problems, they are worth it.
Most day-to-day work is different. Agents, extraction, classification, summarization, RAG, and routine coding need a dependable model with room to run, not maximum intelligence.
That is what Cabrio provides: Qwen3.6-36B at a flat monthly price, with a quota measured in billions of tokens. Around 5 to 10 times cheaper than buying the same volume from a per-token API, and none of the maintenance of self-hosting. When a job needs a stronger model, the DeepSeek V4 Flash plan covers it.
Run your agents 24/7
Point a personal agent like Hermes or OpenClaw, a coding agent, or a folder of cron jobs at Cabrio and let them run. Automated and agentic usage is explicitly allowed.
There are no usage windows, no cooldowns, and no surprise bills. Two concurrent requests, around the clock, all month.
- personal agents
- coding agents
- background automation
- RAG pipelines
- extraction
- classification
- summarization
- cron jobs
- app backends
OpenAI compatible
Change the base URL and keep working. Any OpenAI client, SDK, or agent framework. Streaming, usage metadata, and a stable model ID are included.
from openai import OpenAI
client = OpenAI(
# the only line that changes:
base_url="https://api.cabrioai.com/v1",
api_key=CABRIO_API_KEY,
)Three plans, two models
Qwen3.6-36B in two quota sizes, or DeepSeek V4 Flash for heavier work. Everything else is identical.
- Model
- Qwen3.6-36B
- Input tokens
- 3B / month
- Output tokens
- 50M / month
- Concurrent requests
- 2
- Streaming
- Included
- Model
- Qwen3.6-36B
- Input tokens
- 5B / month
- Output tokens
- 100M / month
- Concurrent requests
- 2
- Streaming
- Included
- Model
- DeepSeek V4 Flash
- Input tokens
- 5B / month
- Output tokens
- 100M / month
- Concurrent requests
- 2
- Streaming
- Included
Quotas reset with each billing cycle. There is no overage billing: when a quota is used up, requests pause until the next cycle. Cancel anytime.
Privately hosted
Cabrio runs on dedicated infrastructure that we operate. Your requests go to our servers and nowhere else.
Your prompts and outputs are never used for training and never read. We count tokens to enforce quotas, and that is it.