chapter11-task-horizon.csv

# METR 50%-success task horizon by model
# Derived from: data/source-receipts/sources/metr-time-horizon.yaml (retained raw copy)
# Upstream: METR, task-horizon evaluation. Benchmark METR-Horizon-v1.1,
#   long_tasks_version 799cc9c4b4483a93fc3445623a49ea1bd74fdeb2 (recorded in the raw file)
# Verified 2026-09-03: the raw file holds 26 model results; this receipt has 26
#   rows, and all 26 p50_horizon_minutes values and all 26 release_date values
#   match the raw metrics.p50_horizon_length.estimate entries exactly.
model,release_date,p50_horizon_minutes
gpt2,2019-02-14,0.053778
davinci_002,2020-05-28,0.144057
gpt_3_5_turbo_instruct,2022-03-15,0.599247
gpt_4,2023-03-14,3.987428
gpt_4_1106_inspect,2023-11-06,4.044959
claude_3_opus_inspect,2024-03-04,3.952262
gpt_4_turbo_inspect,2024-04-09,3.732787
gpt_4o_inspect,2024-05-13,6.991195
claude_3_5_sonnet_20240620_inspect,2024-06-20,11.395377
o1_preview,2024-09-12,20.326586
claude_3_5_sonnet_20241022_inspect,2024-10-22,20.522872
o1_inspect,2024-12-05,38.831588
claude_3_7_sonnet_inspect,2025-02-24,60.388937
o3_inspect,2025-04-16,119.732634
claude_4_opus_inspect,2025-05-22,100.366123
claude_4_1_opus_inspect,2025-08-05,100.472004
gpt_5_2025_08_07_inspect,2025-08-07,203.012577
gemini_3_pro,2025-11-18,224.325884
gpt_5_1_codex_max_inspect,2025-11-19,223.714694
claude_opus_4_5_inspect,2025-11-24,292.994594
gpt_5_2,2025-12-11,352.249302
claude_opus_4_6_inspect,2026-02-05,718.80683
gpt_5_3_codex,2026-02-05,349.530732
gemini_3_1_pro,2026-02-19,384.147435
gpt_5_4,2026-03-05,341.735276
claude_mythos_preview_early_inspect,2026-04-07,1044.780145