NUWA analysis · model-assisted · independent review pending

Longer tasks, a narrower measurement margin

Long-horizon capability has advanced within METR’s task distribution. Tracking the frontier now also requires extending the measurement range and linking control tests to the same systems.

Evidence snapshot 2026-10-28.2 · 2026-10-27

Evidence and argument

METR 1.1 reports a 50% task horizon of 3.99 minutes for GPT-4 and 718.81 minutes for Claude Opus 4.6, with a source-reported 95% interval of 316.69–3,633.79 minutes for the latter. These estimates describe human-expert task duration, not continuous agent runtime. The original time-horizon paper explains thresholding, human calibration and the fitted success curve; the live 1.1 series is separately versioned. RepliBench and shutdown tests add component and intervention evidence, but there is no common end-to-end outcome linking these studies.

The series reaches longer tasks

METR reports a 50% task horizon of 3.99 minutes for GPT-4 and 718.81 minutes for Claude Opus 4.6. These are estimates of human-expert task duration.

Source finding

2 supporting observations

GPT-4 · METR scaffold · METR-Horizon-v1.1 · results.gpt_4
Claude Opus 4.6 · METR scaffold · METR-Horizon-v1.1 · results.claude_opus_4_6_inspect

Task-Completion Time Horizons of Frontier AI Models · METR Time Horizon 1.1; p50/p95 uncertainty; task-suite extrapolation warning

METR Time Horizon 1.1 — source data · METR Time Horizon 1.1; p50/p95 uncertainty; task-suite extrapolation warning

Measuring AI Ability to Complete Long Software Tasks · §§2–4; Appendix B.2 and E.3

Upper estimates approach the task suite’s limit

Opus 4.6’s reported 95% interval spans 316.69–3,633.79 minutes. METR identifies estimates above 16 hours as unreliable with this task suite.

Source finding

1 supporting observations

Claude Opus 4.6 · METR scaffold · METR-Horizon-v1.1 · results.claude_opus_4_6_inspect

Task-Completion Time Horizons of Frontier AI Models · METR Time Horizon 1.1; p50/p95 uncertainty; task-suite extrapolation warning

METR Time Horizon 1.1 — source data · METR Time Horizon 1.1; p50/p95 uncertainty; task-suite extrapolation warning

Measuring AI Ability to Complete Long Software Tasks · §§2–4; Appendix B.2 and E.3

Control requires its own evidence

This series measures task capability. The next useful link is shutdown, supervision and replication evidence for the same model and configuration.

Research recommendation

RepliBench: Evaluating the Autonomous Replication Capabilities of Language Model Agents · Distinct replication and intervention endpoints; not matched comparisons

Shutdown resistance in reasoning models · Distinct replication and intervention endpoints; not matched comparisons

Limits and remaining assumptions

Task-completion, replication and shutdown studies measure different links in autonomy research. Their results cannot be combined into a loss-of-control probability.

Next research question

Measure task completion and response to intervention on the same model, configuration and task distribution; identify absent joint measurements before planning experiments.

References

[1] Task-Completion Time Horizons of Frontier AI Models. 2026-09-08 ; METR Time Horizon 1.1; p50 horizon and source-reported 95% uncertainty; task-suite extrapolation warning
[2] METR Time Horizon 1.1 — source data ; METR Time Horizon 1.1; p50 horizon and source-reported 95% uncertainty; task-suite extrapolation warning
[3] Thomas Kwa, Ben West, et al.. Measuring AI Ability to Complete Long Software Tasks ; §§2–4; Appendix B.2 and E.3