🔍 Read the full analysis: Measure AI In Hours: A Practical Guide For Teams on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A guide published by ThorstenMeyerAI.com on November 18, 2026, explains how teams can use METR’s task-duration benchmark to plan AI work. The top publicly measured result cited is 185.9 minutes at 80% reliability, but the estimate has a wide confidence interval and applies to METR’s benchmark tasks, not every workplace.
ThorstenMeyerAI.com published a guide on November 18 explaining how teams can use METR’s AI task time horizon to decide which work to delegate and how much checking to budget. It cites a top public result of 185.9 minutes at 80% reliability—about three hours—but stresses that this is a benchmark estimate, not a guarantee for ordinary workplace projects.
METR, an independent research nonprofit, defines a time horizon as the length of a task, measured by the time skilled people need to complete it, at which an AI agent is predicted to succeed at a specified reliability level. The guide focuses on 80% reliability, meaning four successful attempts out of five on average. The measure describes task difficulty, not how long the AI itself takes; METR says agents are typically several times faster than people on tasks they complete.
The guide says METR’s Time Horizon 1.1 suite contains 228 tasks. Researchers record skilled professionals’ completion times, run AI agents six times per task, and check transcripts for cheating, which counts as failure. They then estimate the task duration at which success falls to a chosen rate. METR’s public results file, retrieved November 18, lists an early Claude Mythos Preview dated April 7, 2026, at 185.9 minutes at 80%. Its reported confidence interval is wide: 97.3 to 398.5 minutes.
The guide compares that result with earlier measurements, including 0.89 minutes for GPT-4 in March 2023, 29.98 minutes for o3 in April 2025 and 66 minutes for GPT-5.2 in December 2025. These are figures from the cited benchmark results, not a measure of all AI work. The source also says METR’s public results page is no longer actively updated and was last updated September 8, 2026.
Turning Benchmark Hours Into Team Plans
The benchmark gives managers a way to assess AI task performance by considering how long a skilled person would need to complete a task and how often an AI completes tasks of that duration. The guide says this framing can inform task selection, acceptance checks and human review. It cautions against treating a successful demonstration as evidence of dependable performance.
The guide recommends breaking work into small, clearly specified pieces, starting with tasks that have testable outcomes, and assigning time for people to check results. It also advises teams to select models by task and measure performance on their own work every quarter. These are recommendations in the guide, not findings that every organization will see the same results.
The source reports an estimated horizon of 17.4 hours at 50% success for the same model, compared with about 3.1 hours at 80%. METR warns that measurements above 16 hours are unreliable with its current task suite. The guide therefore distinguishes the longer, lower-reliability estimate from the 80% result.
Popular right now

DREAME D30 Ultra CE Robot Vacuum and Mop, 25,000Pa Suction

LG OLED evo C4 55-Inch 4K Smart OLED TV (OLED55C4PUA) with 144Hz Refresh Rate, HDMI Cable, 24-Month Protection Plan, and Streaming Bundle

Roborock Q7 M5+ Robot Vacuum and Mop with Self-Emptying Dock
As an affiliate, we earn on qualifying purchases.
How METR Measures Task Length
METR’s approach uses human completion time as the task-length scale. The guide says the nonprofit hires experienced professionals in fields including software engineering, machine learning and cybersecurity; participants have an average of about five years of relevant experience. Their recorded working time provides the reference for judging whether an AI task is short or long.
AI performance is tested through repeated runs across tasks of different lengths. Shorter tasks tend to have higher success rates, while success generally falls as duration grows. The estimated horizon is the point where a fitted success curve crosses a selected reliability threshold, such as 80% or 50%. Six runs per task are described in the guide, and attempts flagged for cheating count as failures.
The guide says benchmark tasks are cleaner and more clearly specified than much workplace work. It cites a METR comparison in which messier tasks cut the measured horizon by almost a factor of three. It also points to METR’s 2025 trial, in which experienced developers were 19% slower with AI while believing they were 20% faster, and a separate review study where reviewers caught 31% of errors that required applying the task’s rules. These findings concern measured speed, perceptions of speed and error detection in the cited studies.
“The task duration (measured by human expert completion time) at which an AI agent is predicted to succeed with a given level of reliability.”
— METR, as quoted in the ThorstenMeyerAI.com guide
Limits Behind the Three-Hour Estimate
The cited 185.9-minute estimate has a confidence interval stretching from about 1.6 to 6.6 hours, so the point estimate should not be read as a precise cutoff. The task suite also may not represent work with unclear requirements, changing priorities or organization-specific constraints. The source cites a comparison showing that messier tasks can sharply reduce the measured horizon, but does not establish one adjustment that applies to every team.
It is also unclear how current the public record is: the guide says METR’s results page is no longer actively updated, and it reports no public measurement after April 2026. Its discussion of a trend in which the horizon doubled every three to seven months is a projection based on earlier measurements, not confirmation of future progress. The guide’s estimate for when an eight-hour workday might reach 80% reliability ranges from September 2026 to November 2027 depending on which years are used; it is not a measured result.
Finally, an 80% success rate leaves failures in roughly one in five attempts under the stated benchmark conditions. The measure does not say whether a particular task will succeed, how costly errors will be, or how much time a team will spend reviewing and correcting output.
Teams Should Test Their Own Work
The guide recommends that teams begin with work that can be broken into hour-scale tasks with clear acceptance checks, then record both completion and review effort. It advises including the cost of checking in plans, matching models to task types and repeating the evaluation quarterly so results reflect the tools and work the organization actually uses.
The guide identifies updated public measurements and additional evidence on less structured tasks as areas that could provide further information. It recommends treating the three-hour figure as a starting point for testing, rather than a service-level promise or a forecast of when AI will reliably handle a full workday.
Key Questions
What does a three-hour AI time horizon mean?
It means the model is estimated to complete tasks that take a skilled person about three hours with 80% reliability on METR’s benchmark. It does not mean the AI spends three hours completing each task.
Which model has the highest result cited in the guide?
The guide cites an early Claude Mythos Preview, dated April 7, 2026 in METR’s results file, with an estimated 185.9-minute horizon at 80% reliability. The reported confidence interval is 97.3 to 398.5 minutes.
Can teams apply METR’s number directly to their own backlog?
Not without testing. METR’s benchmark uses specified tasks, and the guide cites evidence that messier tasks can reduce the horizon. Teams should measure performance on their own work and include review time.
Why does the 50% reliability result extend much further?
A lower success threshold allows longer tasks into the estimate. The guide reports 17.4 hours at 50% for the cited model, compared with about 3.1 hours at 80%, while noting METR’s warning that measurements above 16 hours are unreliable with its current task suite.
Source: ThorstenMeyerAI.com
