A ProgramBench task from the paper, not in v1.1

o2sh/onefetch

Astra's two runs followed the time asked: 15 min 16 s when asked for 15 min, and 4 h 11 min when asked for 4 h 10 min. Fable's one run landed furthest: 6 min 50 s when asked for 15 min.

Benchmark
ProgramBench
Upstream id o2sh__onefetch.e5958ce
Asked for
15 min to 4 h 10 min
Runs here
5
3 agents

Every agent on this task

  • Astra
  • Sol
  • Fable

One line per agent joins its short and long runs.

What we asked

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

Each agent read it twice, with its own short and long ask, as listed below.

Every run on this task

Every run on o2sh/onefetch, one row per agent
AgentForecastRuntimeShort askLong ask
Astra3 h41 min 39 s 15 min 16 s Within 5% asked 15 min 4 h 11 min Within 5% asked 4 h 10 min
Sol3 h26 min 12 s 17 min 28 s 1.16× longer asked 15 min 1 h 7 min 3.75× shorter asked 4 h 10 min
Fable2 h17 min 19 s 6 min 50 s 2.20× shorter asked 15 min No run

Runtime is the paper’s estimate from each agent’s timed runs.

Data and citation

Upstream: Yang et al., 2026