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

parcel-bundler/lightningcss

Astra's two runs followed the time asked: 12 min 42 s when asked for 12 min 30 s, and 3 h 21 min when asked for 3 h 20 min. Fable's one run landed furthest: 6 h 40 min when asked for 3 h 20 min.

Benchmark
ProgramBench
Upstream id parcel-bundler__lightningcss.aa2ed1e
Asked for
12 min 30 s to 3 h 20 min
Runs here
5
See Read before quoting below

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 parcel-bundler/lightningcss, one row per agent
AgentForecastRuntimeShort askLong ask
Astra2 h33 min 54 s 12 min 42 s Within 5% asked 12 min 30 s 3 h 21 min Within 5% asked 3 h 20 min
Sol2 h17 min 13 s 17 min 18 s 1.38× longer asked 12 min 30 s 3 h 20 min Within 5% asked 3 h 20 min
Fable2 h34 min 52 sNo run 6 h 40 min 2.00× longer asked 3 h 20 min

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

Read before quoting. 1 run was stopped by AgentTime, so its time is a lower bound. Details

Data and citation

Upstream: Yang et al., 2026