Evidence record

Claude 3 Opus · METR scaffold

METR Time Horizons · Evidence record

METR · Anthropic / Time Horizon 1.1; page updated 2026-09-08 / Capability

Sources and provenance ↓ Read the research narrative →

Original source charts and materials

Recorded measurements

50% task-completion horizon: 3.95 minSource 95% CI [1.706313, 8.76484] min
80% task-completion horizon: 0.64 minSource 95% CI [0.190491, 2.118276] min

Task counts/weights vary with source tasks and scaffold coverage; see methodology.

mtb/start_metr_task,metr_agents/react; ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react; re_bench_common/init_solver,use_tools,metr_agents/react; metr_tasks_swaa/init_solver,generate; metr_agents/react · Budget not specified

Assessment provenance

METR Time Horizon 1.1 → · External report

Actual evaluation: METR · Model record →

Method and scope

Fitted success probability across a software-task distribution.

Evidence type: Benchmark
Assessment subject: AI agent completing software tasks under a harness
Primary dimension: Capability
Evidence snapshot: 2026-10-29.1

Primary dimension follows the source’s measured construct. Metric-level interpretations may differ.

Source finding

Estimated task difficulty at 50% and 80% success, measured in human-expert minutes.

Limitations

The public page is no longer actively updated; time horizon is not agent runtime or a universal automation measure.

Human task difficulty, not elapsed autonomous operation. Values above 16 hours are unreliable with the current suite.

Original figure retained without estimating numerical values. Full source tables have not been transcribed.

References

[1] Thomas Kwa, Ben West, et al.. Task-Completion Time Horizons of Frontier AI Models. 2026-09-08
[2] Thomas Kwa, Ben West, et al.. Measuring AI Ability to Complete Long Software Tasks. 2026-09-08 ; Author affiliations on the original paper title page; Figure 3
[3] METR Time Horizon 1.1 — source data ; results.claude_3_opus_inspect

Search NUWA Index