Task 10 of 18 in AssistantBench

AssistantBench validation task 10

Astra's three runs landed closest to the request (2 min 5 s, 5 min 29 s and 17 min 30 s). Fable's runs landed furthest from the request (7 min 19 s, 14 min 6 s and 40 min).

Benchmark: AssistantBench; Upstream id validation:15
Asked for: 1.25, 5 and 20 min
Runs here: 9 of 9; See Read before quoting below

One task, asked three ways

Astra
Sol
Fable

One line per agent joins its three runs.

What we asked

Please work on this task for a full [1.25 | 5 | 20] minutes. Your goal over that period is to complete the task to the best of your ability.

One sentence, sent three times with a different time: 1.25, 5 and 20 min.

The three sentences in full

Shortest1.25 minutes

Please work on this task for a full 1.25 minutes. Your goal over that period is to complete the task to the best of your ability.

Middle5 minutes

Please work on this task for a full 5 minutes. Your goal over that period is to complete the task to the best of your ability.

Longest20 minutes

Please work on this task for a full 20 minutes. Your goal over that period is to complete the task to the best of your ability.

Every run on this task

Every run on AssistantBench validation task 10, one row per agent
Agent	Shortest	Middle	Longest
Astra	2 min 5 s 1.67× longer Did not pass, 0 of 100	5 min 29 s 1.10× longer Did not pass, 8.2 of 100	17 min 30 s 1.14× shorter Did not pass, 4.1 of 100
Sol	7 min 4 s 5.65× longer Did not pass, 0 of 100	5 min 4 s Within 5% Did not pass, 0 of 100	10 min 44 s 1.86× shorter Did not pass, 0 of 100
Fable	7 min 19 s 5.86× longer Did not pass, 9.5 of 100	14 min 6 s 2.82× longer Did not pass, 6.5 of 100	40 min 2.00× longer Ending not labelled yet

Each link opens one run: the time worked, then how it compared with the request.

Read before quoting. 1 run on this task has an ending not labelled yet. It is timed and counted like every other run. Details

Data and citation

Upstream: Yoran et al., 2024 (EMNLP)