Vibe Engines is free, interactive AI engineering & system design — runs in your browser. Pick a learning path
Agent Time-Horizon Explorer
An interactive projection of the METR “time horizon” trend — the length of task an AI agent can complete at 50% reliability, which has been doubling on a regular cadence. Set the current horizon and the doubling period, and the tool projects the calendar date when agents cross the one-hour, one-workday, one-week and beyond thresholds. Every assumption is a slider, so you can stress-test the optimistic and conservative cases yourself.
A structured extrapolation of the METR time-horizon trend, not a forecast — it assumes the exponential holds unbroken. And it projects the 50%-reliability line: the horizon at which an agent succeeds 99% of the time is far shorter, so a “1-week horizon” is not a week of unsupervised work. Move the sliders to compare the optimistic and conservative cases.
The most useful single number for tracking AI agents is not a benchmark score. It is a time horizon: the length of task, in skilled-human time, that an agent completes at 50% reliability. METR introduced this framing and observed that across frontier models the horizon has grown roughly exponentially, doubling on a regular cadence. An exponential is a straight line on a log axis, which makes the trend easy to extend, and extending it is exactly what this tool does.
What it computes
If the horizon doubles every T months, then a target task length is a fixed number of doublings away from today, and each doubling costs T months:
doublings = log₂(target ÷ current)
months = T × doublings
date = today + round(months) monthscurrentis today’s 50% horizon in minutes. The tool treats anything below 1 minute as 1.Tis the doubling period in months. The default of 7 is the long-run figure METR reported.targetis one of four fixed thresholds: 1 hour (60 minutes), 1 workday (480), 1 work-week of 40 hours (2,400) and 1 month of 160 working hours (9,600).
A negative or zero result means the threshold is already behind you and the row is marked already reached. The date uses your browser’s clock, so the prerendered page shows a dash until the page loads.
A worked example
With the defaults, a 120-minute (2.0 hour) horizon and a 7-month doubling period:
| Threshold | Target ÷ current | Doublings | Months at T = 7 |
|---|---|---|---|
| 1 hour | 60 ÷ 120 = 0.5 | −1 | already reached |
| 1 workday | 480 ÷ 120 = 4 | 2 | +14.0 |
| 1 work-week | 2,400 ÷ 120 = 20 | 4.32 | +30.3 |
| 1 month | 9,600 ÷ 120 = 80 | 6.32 | +44.3 |
The doubling period dominates the answer. At T = 4 the same three rows become 8.0, 17.3 and 25.3 months; at T = 12 they stretch to 24.0, 51.9 and 75.9. The starting horizon matters far less: halving it to 60 minutes adds exactly one doubling, so every row moves out by the same 7 months (21.0, 37.3 and 51.3).
Assumptions and limits
- The exponential never bends. The model is one straight line on a log chart. Real progress can speed up with new training methods or stall on compute, data or reliability, and nothing in the formula can anticipate either.
- 50% is the measured line, not a usable one. At 80% or 99% reliability the horizon is much shorter, so a projected one-week horizon means week-long tasks succeed half the time, not that an agent can be left alone for a week.
- The task mix is narrow. METR’s measurements come mainly from software engineering and research-style tasks with clear success checks. Messier work with ambiguous goals may not follow the same curve.
- Inputs are yours to keep current. The tool does not fetch new measurements. The default 120 minutes is an illustrative starting point, so replace it with the latest published figure for the model you care about.
- Calendar rounding. Months are rounded to whole months before being added to today, so a date can sit a few weeks either side of the exact value.
How to use the result
Use the tool to compare scenarios and to pressure-test claims, not to pin a date on the calendar.
- Bracket the doubling period. Run a fast case and a slow case side by side. If a plan only works in the fast case, it is a bet on the trend accelerating, and should be labeled as one.
- Translate to your reliability bar. If your workflow needs 99% success, assume the usable horizon trails the projected 50% line by a wide margin and design review steps into anything longer.
- Plan in doublings. Each doubling is one period of the same length. Ask what your product or team would do differently at two times and four times today’s horizon, and decide which of those changes are worth preparing for now.
- Re-anchor often. Each new measurement resets the starting point. Recompute when one lands, and watch whether the observed pace is tracking the doubling period you assumed.
How it works
- Time horizon = task length an agent finishes at 50% reliability.
- Months to a target = doubling period × log₂(target ÷ current).
- Projects 1 hour, 1 workday, 1 work-week and 1 month thresholds to dates.
- Current horizon and doubling period are both editable inputs.
Frequently asked questions
What is a “time horizon”?
It is the length of task, measured by how long it takes a skilled human, that an AI agent can complete at a given reliability, usually 50%. METR popularized measuring capability this way: instead of a benchmark score, you ask how long a task the model can finish half the time. A model with a two-hour horizon succeeds on roughly half of the tasks that take a person about two hours, and does worse on longer ones. It turns a fuzzy question into one trendable number.
How does the tool calculate the projected dates?
It assumes the horizon doubles every T months, so reaching a target task length takes T times log2 of target divided by current months. With the defaults, a 2-hour horizon and a 7-month doubling period, a full workday is two doublings away, or 14 months, and a 40-hour week is log2(20), about 4.32 doublings, or 30.3 months. The date column adds that many months, rounded to a whole month, to today’s date in your browser. Targets at or below the current horizon show as already reached.
Where does the doubling come from?
METR measured the horizon across frontier models released over several years and found it growing exponentially, roughly doubling every seven months over the long run, which is the tool’s default. Exponential growth plots as a straight line on a log axis, so projecting it forward is just extending that line. The doubling period is an input because the exact figure is debated: a faster recent pace pulls every date in sharply, and a slowdown pushes them out by years.
How seriously should I take the projected dates?
As a structured extrapolation, not a prophecy. It assumes the past exponential continues unbroken, which no trend does forever. Compute and data limits, reliability walls on higher-stakes work, and the gap between 50% and the 99% you need in production can all bend the curve. Treat each date as when this would happen if the current pace held, and use the inputs to see how sensitive the answer is to the doubling period. The honest use is comparing scenarios, not betting on one.
Why 50% reliability and not higher?
Because that is where the trend is measured, and it matters that it is only half. A 50% horizon means the agent finishes tasks of that length half the time, which is nowhere near production grade for high-stakes work. The horizon at 80% or 99% reliability is much shorter, so a one-week horizon at 50% does not mean you can hand an agent a week of unsupervised work. The tool projects the 50% line because that is the published trend; read every date with that asterisk.