🔍 Read the full analysis: AI’s Time Horizon, Explained: What Teams Can Expect Today on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
METR’s public benchmark data, as reported by ThorstenMeyerAI on November 18, 2026, gives an early Claude Mythos Preview an 80% time horizon of 185.9 minutes on its tasks. That is a measure of human-rated task length, not how long the AI works, and the result may not transfer to messy workplace projects. Public measurements after April 2026 were not available in the source.
A November 18 guide from ThorstenMeyerAI uses METR’s public benchmark to give teams a practical measure of the work AI agents can handle: an early Claude Mythos Preview reached an estimated 80% success rate on tasks taking skilled people about 185.9 minutes. The figure is a benchmark estimate, not a guarantee for workplace projects, and the guide says METR’s public results have not been updated with measurements after April 2026.
METR, an independent research nonprofit, defines an AI agent’s time horizon by the duration of tasks, measured using human expert completion time, at which the agent is predicted to succeed at a chosen reliability level. It is not the amount of time the AI spends working. According to the guide, METR’s Time Horizon 1.1 suite contains 228 tasks spanning seconds to more than eight hours. Skilled professionals complete the tasks to establish human-time estimates; AI systems then make six independent attempts per task, with cheating counted as failure.
The guide reports that METR’s results file lists an 80% horizon of 185.9 minutes for an early Claude Mythos Preview dated April 7, 2026. The estimate’s confidence interval is wide: 97.3 to 398.5 minutes, or roughly 1.6 to 6.6 hours. The results file records earlier milestones including 29.98 minutes for o3 in April 2025 and 66 minutes for GPT-5.2 in December 2025. These are benchmark results, not independent evidence that every model will perform similarly on a company’s own work.
Reliability changes the result. For the same Mythos Preview, the guide cites a 50% horizon of 1,044.8 minutes, about 17.4 hours, compared with about 3.1 hours at 80%. METR cautions that estimates above 16 hours are unreliable with its current task suite. The guide also cites METR findings that messier tasks can reduce the horizon by almost a factor of three. Those caveats limit how directly teams can map the headline number onto real projects.
Turning Benchmarks Into Work Plans
The measure gives managers a more useful planning question than whether AI can complete a particular task at all: how long would a skilled person need, and how often does the agent finish comparable work? That can help teams decide which tasks to delegate, where to split projects into smaller units, and how much review time to budget.
The 80% result also makes the cost of oversight visible. Four successful attempts out of five leaves a meaningful failure rate; teams need a person or a reliable test to catch errors before they affect customers or systems. The source guide cites a 2025 METR trial in which experienced developers were 19% slower while using AI, despite believing they were 20% faster. It also cites a separate study where reviewers caught 31% of errors that required applying task rules. Those findings argue against treating a benchmark score as a substitute for workflow evaluation.
For practical use, the guide recommends breaking work into hour-scale pieces with clear acceptance checks, trying AI first on testable tasks, and tracking review effort. Teams should measure performance on their own work rather than assuming benchmark results carry over unchanged.
Products Worth Considering

Samsung 27-Inch Odyssey G5 QHD Gaming Monitor (G50F)

Jackery HomePower 3000 Portable Power Station, 3072Wh

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, White Smoke
As an affiliate, we earn on qualifying purchases.
How METR Measures Task Length
METR’s method compares AI attempts with work completed by experienced people. The guide says the human contractors average about five years of relevant experience in areas including software engineering, machine learning and cybersecurity. Human completion time gives each benchmark task its duration; AI attempts are then assessed for successful completion, with transcripts reviewed for cheating.
METR fits a curve relating task duration to success rate. The estimated task duration at the point where the curve crosses 80% is the stricter planning horizon; the 50% horizon reflects a lower reliability threshold. The guide notes that an 80% horizon does not mean the model succeeds on every task of that exact length, and shorter or longer individual tasks can produce different outcomes.
The source says METR’s public results page was last updated September 8, 2026, and is no longer actively updated. It also cites a February–March 2026 Frontier Risk Report estimating that labs’ internal frontier was at least three hours but under four hours at 80% reliability. That estimate is distinct from the public benchmark record and should not be read as a current, independently verified measurement.
“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 guide
Where the Three-Hour Estimate Breaks
The public benchmark result is an estimate with a broad confidence interval, and the source reports no public measurements later than April 2026. METR’s page was last updated in September 2026, according to the guide, so the reported record should not be presented as a measurement of the newest systems available on November 18.
It is also unclear how well METR’s tasks represent a particular company’s code, tools, documentation, security rules or approval processes. The source says less tidy tasks can shorten the horizon substantially, but does not provide enough detail here to establish how much any specific team should discount its own work. The benchmark’s 80% figure is not a guarantee for an individual task, and its longer-horizon estimates have additional reliability limits.
The guide offers projections that an eight-hour task horizon at 80% could arrive between September 2026 and November 2027, depending on the years used in the trend fit. That is a conditional forecast, not a measured result. Whether the trend continues, and whether benchmark gains translate to dependable work in operational settings, remain unresolved.
Testing AI Against Team Backlogs
For teams deciding what to automate now, the source guide recommends starting with small, clearly specified tasks and an acceptance test that can establish whether the work is correct. Managers can record completion and error rates, time spent on human review, and whether failures are easy to detect. Those measurements can show whether a tool saves time in a particular workflow, rather than relying on a general benchmark.
The guide also recommends reassessing performance every quarter and choosing models by task. That cadence matters because public benchmark data may lag behind model releases, while a team’s own tools and processes can change. No new METR public measurement date or next benchmark release is specified in the source, so a further public update cannot be assumed.
Key Questions
What does an AI time horizon measure?
It estimates the human working-time length of tasks on which an AI agent reaches a chosen success rate. In METR’s 80% measure, the estimate is where the fitted success curve reaches four successes in five.
Does a three-hour horizon mean the AI works for three hours?
No. It refers to the length of the task as measured by skilled human completion time, not the duration of the AI’s work. The guide says agents can complete tasks they succeed at several times faster than people.
Which model has the latest result cited in the guide?
The guide cites an early Claude Mythos Preview, dated April 7, 2026 in METR’s results file, at 185.9 minutes for the 80% horizon. Its cited confidence interval runs from 97.3 to 398.5 minutes.
Can teams use the benchmark to assign whole projects?
Not by itself. METR’s result comes from benchmark tasks, and the guide says messier tasks can have shorter horizons. Teams should split work into testable pieces and measure success and review costs on their own workflows.
Is an eight-hour AI task horizon confirmed?
No. The guide gives a forecast range of September 2026 to November 2027 based on different trend fits. It is a projection, not a publicly measured result, and the source says public measurements after April 2026 were unavailable.
Source: ThorstenMeyerAI.com
