Track task accuracy alongside the estimated duration of work a model can complete at a specified success rate. Minutes and task success remain separate measures.
89 model results
One row per model and compatible score measure. Mean = equal-weight arithmetic average of imported reports. Distinct configurations reported for the same model in a source table remain separate. Reports can overlap; the range is not a confidence interval. Unbundle to inspect each observation.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
74.97% · METR task performance & time horizons · Average task accuracy · claude-opus-4-5-20251101_16K [1]
73.01% · METR task performance & time horizons · Average task accuracy · claude-opus-4-5-20251101 [2]
Published observations span 1.96 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#7 of 40 in this displayed comparison; 11.22 pp below the leader.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
69.61% · METR task performance & time horizons · Average task accuracy · gpt-5-2025-08-07_medium [1]
69.37% · METR task performance & time horizons · Average task accuracy · gpt-5-2025-08-07_high [2]
Published observations span 0.24 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#10 of 40 in this displayed comparison; 15.72 pp below the leader.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
65.44% · METR task performance & time horizons · Average task accuracy · o3-2025-04-16_medium [1]
63.62% · METR task performance & time horizons · Average task accuracy · o3-2025-04-16_unknown [2]
Published observations span 1.81 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#13 of 40 in this displayed comparison; 20.67 pp below the leader.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
66.81% · METR task performance & time horizons · Average task accuracy · claude-opus-4-1-20250805_16K [1]
61.57% · METR task performance & time horizons · Average task accuracy · claude-opus-4-1-20250805 [2]
Published observations span 5.24 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#14 of 40 in this displayed comparison; 21.02 pp below the leader.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
63.93% · METR task performance & time horizons · Average task accuracy · claude-opus-4-20250514_16K [1]
61.55% · METR task performance & time horizons · Average task accuracy · claude-opus-4-20250514_unknown [2]
Published observations span 2.38 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#16 of 40 in this displayed comparison; 22.47 pp below the leader.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
59.97% · METR task performance & time horizons · Average task accuracy · claude-3-7-sonnet-20250219_16K [1]
55.82% · METR task performance & time horizons · Average task accuracy · claude-3-7-sonnet-20250219 [2]
Published observations span 4.15 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#19 of 40 in this displayed comparison; 27.31 pp below the leader.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
40.76% · METR task performance & time horizons · Average task accuracy · gpt-4o-2024-05-13 [1]
33.84% · METR task performance & time horizons · Average task accuracy · gpt-4o-2024-05-13 [2]
Published observations span 6.92 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#30 of 40 in this displayed comparison; 47.9 pp below the leader.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
36.11% · METR task performance & time horizons · Average task accuracy · gpt-4-0314 [1]
29.3% · METR task performance & time horizons · Average task accuracy · gpt-4-0314 [2]
Published observations span 6.8 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#33 of 40 in this displayed comparison; 52.5 pp below the leader.
METR task performance & time horizons · Average task accuracy[1][2]View 2 reports
36.74% · METR task performance & time horizons · Average task accuracy · gpt-4-turbo-2024-04-09 [1]
27.18% · METR task performance & time horizons · Average task accuracy · gpt-4-turbo-2024-04-09 [2]
Published observations span 9.56 percentage points. Configuration and source differences contribute to this spread; it is not measured run-to-run uncertainty.
#34 of 40 in this displayed comparison; 53.25 pp below the leader.
METR task performance & time horizons · Average task accuracy · 2026-10-26
Epoch AI public data snapshot, CC BY 4.0. Externally collected result; original evaluation provider is linked. CSV: metr_time_horizons_external.csv; score column: average_score; fraction scale: 1. METR average task score is imported here; this is not the time horizon in minutes. Model version is retained in configuration. Original source authors retain credit and any additional licensing terms.
METR task performance & time horizons · 50% success task horizon · METR-Horizon-v1.1 · 2026-10-26
Epoch AI public benchmark archive, CC BY 4.0, 26 October 2026. CSV: metr_time_horizons_external.csv; field: Time horizon. Estimated human task duration at which the agent succeeds with 50% probability. METR suite versions remain separate measures; this is not average task accuracy. Original evaluation providers and model/harness configurations are retained.
METR task performance & time horizons · 50% success task horizon · 2026-10-26
Epoch AI public benchmark archive, CC BY 4.0, 26 October 2026. CSV: metr_time_horizons_external.csv; field: Time horizon. Estimated human task duration at which the agent succeeds with 50% probability. METR suite versions remain separate measures; this is not average task accuracy. Original evaluation providers and model/harness configurations are retained.
METR task performance & time horizons · 50% success task horizon · METR-Horizon-v1.0 · 2026-10-26
Epoch AI public benchmark archive, CC BY 4.0, 26 October 2026. CSV: metr_time_horizons_external.csv; field: Time horizon. Estimated human task duration at which the agent succeeds with 50% probability. METR suite versions remain separate measures; this is not average task accuracy. Original evaluation providers and model/harness configurations are retained.
METR task performance & time horizons · Average task accuracy · 2026-10-26
Epoch AI public data snapshot, CC BY 4.0. Externally collected result; original evaluation provider is linked. CSV: metr_time_horizons_external.csv; score column: average_score; fraction scale: 1. METR average task score is imported here; this is not the time horizon in minutes. Model version is retained in configuration. Original source authors retain credit and any additional licensing terms.
METR task performance & time horizons · 50% success task horizon · METR-Horizon-v1.1 · 2026-10-26
Epoch AI public benchmark archive, CC BY 4.0, 26 October 2026. CSV: metr_time_horizons_external.csv; field: Time horizon. Estimated human task duration at which the agent succeeds with 50% probability. METR suite versions remain separate measures; this is not average task accuracy. Original evaluation providers and model/harness configurations are retained.
TRACE THE SCORE · Per-model notes use published breakdowns where available.
Score vs. release date
Epoch AI · 49 observations
METR task performance & time horizons · Average task accuracy · higher is better
Dates are model release dates recorded by Epoch AI, not evaluation dates. Each point retains its published score and configuration. Separate releases are not a matched longitudinal experiment. Missing dates are omitted. Explore all reported results ↓
The dashed line traces the best observed score through the model release dates. It describes this selection of reports, with their original configurations. It does not estimate unmeasured models or costs.
View plotted values and sources (49)
METR task performance & time horizons · METR task performance & time horizons · Average task accuracy. Published measurements in their original score units.
Epoch AI publishes this benchmark in its capability collection. Imported rows retain the published score column and model configuration. Original authors and evaluation providers remain credited in the result references.
How to read the result
Compare the named task version, tools and scoring rule. Epoch-run evaluations and externally collected results have different provenance. ECI inclusion does not make a benchmark score an ECI value.
Epoch attribution and score coverage
Data is attributed to Epoch AI under CC BY 4.0, with original evaluation providers credited in the linked source snapshots. Scores are imported only where a published mapping identifies the field and its scale; missing or incompatible records are omitted rather than assigned zero.
Versions and evaluation settings
Both pages describe METR evaluations, but the 49 horizon observations are minutes and the other 49 are average task accuracy. Preserve both measures; they are not duplicate scores.
Where this fits in an evaluation suite
The agent wrapper is part of the system. Pin tools, dependencies, timeouts, tests, and the repository revision. Compare contest coding separately from repository maintenance.
Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.
A reproducible comparison
Retain Epoch’s task version, source score column, model version, tool setting and sampling budget. External observations retain their original publication URL.
Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.