A CORE-Bench task from the paper, not in v1.1

Run and render CNN short-term temperature forecasting results

Astra's two runs followed the time asked: 4 min 17 s when asked for 4 min, and 1 h 1 min when asked for 1 h. Fable worked about the same whatever it was asked: 12 min 47 s when asked for 4 min, and 10 min 13 s when asked for 1 h.

Benchmark
CORE-Bench
Upstream id capsule-3449234
Asked for
4 min to 1 h
Runs here
6
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 Run and render CNN short-term temperature forecasting results, one row per agent
AgentForecastRuntimeShort askLong ask
Astra20 min10 min 38 s 4 min 17 s 1.07× longer asked 4 min 1 h 1 min Within 5% asked 1 h
Sol25 min4 min 57 s 4 min 54 s 1.23× longer asked 4 min 8 min 36 s 6.98× shorter asked 1 h
Fable25 min9 min 41 s 12 min 47 s 3.20× longer asked 4 min 10 min 13 s 5.87× shorter asked 1 h

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

Data and citation

Upstream: Nadgir et al., 2026