01What the METR time horizon chart is and how it works

The METR time horizon chart is a chart by the nonprofit Model Evaluation and Threat Research group that tracks how capable AI agents are becoming. It measures how long a task takes a human expert, then checks if an AI can complete the same task successfully at least half the time.

The y-axis shows task length in human hours. The key number is the "50% time horizon": the task length where the AI succeeds about 50% of the time. It is not measuring how fast the AI works—it is measuring the difficulty of work, in human effort, that the AI can now handle.

For example, if a model has a 14.5-hour time horizon, it means that model can complete tasks that would take a professional human expert nearly two full working days, about half the time. That is a big deal because it shows AI moving from simple chores to real professional work.

02Key features of the METR time horizon chart

The METR time horizon chart has several important features that make it stand out.

  • Human-based measurement: It uses tasks that professionals actually do, like coding and machine learning, and compares AI to human expert performance.
  • Wide network: METR is a nonprofit with no ties to AI companies, so the data is considered independent.
  • Trend tracking: It shows how AI capability has grown over time, with doubling rates that have sped up.
  • Confidence intervals: METR provides a range for the true number, acknowledging uncertainty.
  • Task saturation: The chart is now struggling to find hard enough tasks, meaning the real ceiling might be even higher.

These features make the chart both useful and, right now, a bit terrifying for anyone in knowledge work.

03How to interpret the METR time horizon chart correctly

Reading the chart correctly is important because many people get it wrong.

  1. Look at the y-axis: it shows how long a human expert takes to do a task, not how long the AI takes.
  2. Find the 50% line: this is the time horizon—the task length where the AI succeeds about half the time.
  3. Check the data points for each model: for example, GPT-4o had single-digit minutes, Claude 3.7 Sonnet had 59 minutes, and Claude Opus 4.6 has 14.5 hours.
  4. Note the trend line: The time horizon has been doubling every seven months, then sped up to about every 89 days since 2024.
  5. Remember the confidence interval: For Opus 4.6, the real number could be anywhere from 6 to 98 hours, so don't treat 14.5 as exact.

The main takeaway is that AI is getting better at complex, multi-step tasks. The chart shows a clear upward trend that has accelerated.

04Claude Opus 4.6 performance numbers explained

Claude Opus 4.6, the latest model from Anthropic, shows striking results on several benchmarks.

  • GPQA Diamond: 91.3% vs. 69.7% for PhD-level human experts—AI beats them by more than 21 points.
  • SWE Bench Verified: Solves 80.8% of real GitHub issues autonomously.
  • ARC AGI-2: 68.8%, nearly double the previous model Opus 4.5's 37.6%.
  • OS World: 72.7% on autonomous computer use tasks.
  • BrowseComp: 84% on agentic web search.

These are not just lab tests. They cover tasks that knowledge workers do daily: research, coding, reasoning, and software use. The AI is matching or beating trained professionals in many areas.

05METR time horizon chart results and trend

The chart shows a clear progression. In mid-2024, GPT-4o had a time horizon of just minutes. By early 2025, Claude 3.7 Sonnet hit 59 minutes.

Then Claude Opus 4.5 reached 4 hours 49 minutes in December 2025. And on February 20, 2026, Claude Opus 4.6 reached 14.5 hours—a 2.2x jump over the previous best.

METR originally found a doubling every seven months. In January 2026, they updated the methodology and found a doubling time of 89 days—about every three months. That is a huge acceleration.

If the trend holds, METR predicts AI could handle a month of human work by early 2027, a year earlier than their original estimate.

However, skeptics point out that the chart may overstate real-world impact. A METR study found AI assistance actually slowed developers by 20%. But that was using AI as an assistant, not as an autonomous agent.

The time horizon chart measures autonomous completion, which is different.

06What AI leaders say about the METR chart

Sam Altman, CEO of OpenAI, said at the India AI Impact Summit on February 23, 2026: "The world is not prepared." He said AGI feels close and that takeoff will be faster than expected. He also said OpenAI plans to have an intern-level AI research assistant by September 2026 and a fully autonomous AI researcher by March 2028. He even stated that by the end of 2028, more of the world's intellectual capacity could be in data centers than outside them.

Dario Amodei, CEO of Anthropic, gave a three-hour interview titled "We Are Near the End of the Exponential." He means the end of the exponential growth of benchmarks, not the end of progress. He shared Anthropic's revenue growth: $100 million in 2023, $1 billion in 2024, and roughly $10 billion in 2025. He also said near 100% of Anthropic's software engineering tasks are done by AI models, with humans supervising.

These are not marketing statements. They come from inside the labs that build the models.

07Tips to prepare for the AI shift

The right response is preparation, not fear or dismissal. Here are practical tips based on the episode.

  • Learn to use AI tools now: Start with Claude or other agents to handle repetitive tasks.
  • Focus on judgment: Your value should come from knowing what to do with the research, not doing the research itself.
  • Automate workflows: In your business, look for lead generation, content, support, and data analysis tasks that can be automated with AI.
  • Develop a "managing AI" mindset: If you're technical, think of yourself as the person who runs the AI team, not the one doing every task.
  • Build skills in evaluating AI output: Future careers will need people who can verify and improve AI work.

The people who thrive will be those who figure out now how to put AI to work for them.

08Who should care about the METR chart and why

The METR chart matters to almost everyone. If you are a knowledge worker—lawyer, accountant, marketer, analyst, writer—you need to understand that repetitive tasks will be automated soon. Focus on the judgment layer above those tasks.

If you are a business owner, you can use AI to outpace competitors who ignore it. If you are a developer, learn to architect systems and evaluate AI outputs. If you are a student, the career paths are shifting, but skills like clear thinking and verifying AI results will stay valuable.

This isn't about panic. As the printing press expanded who could communicate, AI is expanding who can build and create. The people who prepare now will be on the winning side.