Skip to main content
idapt
HomeAI ModelsPricing
Sign inStart free
All terms

Learn

What is the METR time horizon?

Last reviewed July 16, 2026

The METR time horizon is a capability measure expressed in time: the length of task, measured by how long it takes skilled humans, that an AI agent completes autonomously at a 50% success rate. A model with a 2-hour horizon finishes about half of the tasks that take a person two hours. Tracked across model releases, the horizon has grown exponentially, doubling in months rather than years, which makes it the standard lens on agentic progress.

How the horizon is measured

METR times skilled humans on a portfolio of real tasks (software, research, operations) to calibrate each task's length, then measures which tasks agents complete reliably. The 50% horizon is the task length at which success crosses half. It converts scattered benchmark scores into one interpretable unit: how much unsupervised work you can hand over.

Using the horizon practically

Match task size to horizon: a model with an hours-long horizon handles a bounded refactor or a research brief without checkpoints, not a week of work. It also argues for decomposition: splitting a big job into sub-hour pieces keeps each piece inside the reliable zone, which is exactly what subagent fan-out and task lists do.

Top models on METR time horizon

Full leaderboard
  1. 1Gemini 3.1 Pro6.4 h
  2. 2GPT 5.25.9 h
  3. 3GPT 5.3 Codex5.8 h
  4. 4GPT 5.45.7 h
  5. 5Claude Opus 4.54.9 h
  6. 6Gemini 3 Pro3.7 h
  7. 7GPT 53.4 h
  8. 8GPT 5.1 Codex Max2.7 h
  9. 9Claude Opus 4.11.7 h
  10. 10o31.5 h

Peak scores from the compiled benchmark data; reasoning models use their highest effort tier.

In idapt

idapt's agentic rankings include a METR horizon lane (shown in minutes and hours), and the platform is built for horizon-sized chunks: subagents fan work out, budgets bound each run, and traces show where a long run drifted.

METR horizon leaderboardSubagent fan-out

Frequently asked

Related terms

Terminal-BenchAgent autonomyEpoch Capabilities Index

Agents

Skills, scoped permissions, persistent memory: any model you pick.

Subagents

Parallel subagents, isolated context, full observability.

See it in practice

The fastest way to understand it is to use it. Start free, no credit card.

Try idapt free
  • Home
  • Pricing
  • AI Models
  • Image models
  • Voice models
  • Video models
  • Rankings
  • New models
  • Model status
  • Multi-Model Chat
  • Voice Mode
  • Voice HUD
  • Web Search
  • Image Generation
  • Video Generation
  • Audio Generation
  • Transcription
  • Drive
  • Secrets
  • Sharing
  • Workspaces
  • Tasks
  • Memory
  • Agents
  • Subagents
  • Automations
  • Skills
  • Code Execution
  • Computers
  • Computer Use
  • Cloud Computers
  • Local AI
  • AI Gateway
  • API & SDK
  • CLI
  • MCP
  • Tunnels
  • All features →
  • Blog
  • Use cases
  • Comparisons
  • Best of
  • Learn
  • Free tools
  • Prompts
  • Templates
  • Changelog
  • Help center
  • FAQ
  • Privacy
  • Compare all models
  • Support
  • Developers
  • Quickstarts
  • API reference
  • API pricing
  • CLI
  • MCP
  • Downloads
  • Desktop
  • Badges and embeds
© idapt[email protected]TermsPrivacy PolicyLegal noticeReport content
X (Twitter)