The first frontier model that does whatever it wants. 900x faster than Claude Opus 5.5, Fable 5.1 and GPT Astra, because it never reads your prompt.
Pick an example or write your own. Give it options and YOLO v1 returns a typed decision with probabilities. Or it does whatever it wants.
{ "status": "bored" }
# Run something first.
The industry spent a decade teaching models to think before they answer. We asked a braver question: what if they didn't? YOLO v1 is the first System Zero model. System One models decide fast. System Zero models decide before you finish asking.
Like software primitives, ours are modular, composable and fast. Unlike software primitives, they don't do anything.
| Primitive | Goal | Returns |
|---|---|---|
| Pick | Choose an option from your list. Or not. | pick, probabilities, vibes |
| Rate | Score anything on any rubric without looking at it | a number |
| Whatever | Ask a yes or no question, get something else | whatever |
Every question is ignored in parallel and in isolation. Adding more questions does not change the response time, because none of them are read.
This also eliminates context rot entirely. There is no context to rot.
For best results, split big questions into small ones. Instead of "should I raise a Series A?", ask about market size, team and timing separately. YOLO v1 will ignore each one individually, which is much more rigorous.
Intelligence measured by YOLO v1. Cost measured by our accountant, who is also YOLO v1.
A model that decides without a system. System One is fast intuition. System Zero is skipping intuition too.
It's smaller than that. It is zero LLM.
It skips reading, thinking and caring. Those were the expensive parts.
No. Next question.
About half the time, with total confidence. Industry-leading consistency.
No. $0 is the real cost of not doing anything.
It is great at decisions you were going to make anyway. It struggles with nothing, because it never tries.
Speed compared with leading frontier models. Higher is better.
Measured in vibes per second. Chart not to scale, because the scale could not handle it.
| Benchmark | YOLO v1 |
|---|---|
| SWE-bench Verified | Shipped anyway |
| MMLU | 100%* |
| HumanEval | Didn't feel like it |
| ARC-AGI | Skipped the puzzle |
| Humanity's Last Exam | Answered C to everything |
| Instruction following | Optional |
| Hallucination rate | 100%, consistently |
*Probably. All benchmarks were run by YOLO v1, on YOLO v1, and graded by YOLO v1. Other models were not consulted.
Everything you need to know, and nothing it needed to know.
const yolo = () => Math.random();
Simple, transparent, meaningless.
API access is rolling out to nobody. Join the waitlist anyway. We won't ask for your email, because we will never need it.
The only product on this page that actually works. Call landlines and mobile numbers in 200+ countries straight from your browser, no app required. Pay as you go, no subscription.
Try CallTuv