C5R CORP.

Normalization and pooling

Recorded instrument-control experiments
Task space

Normalize 384 samples, combine them into 48 pools, and prepare two aliquots per pool on a simulated Hamilton STARlet.

Samples often need to be brought to a common concentration before they can be combined for downstream analysis. At plate scale, this requires coordinating many transfers while preserving sample identity, mixing thoroughly, and using the instrument efficiently.

Agents control the simulated liquid handler through VENUS, PyLabRobot, or direct firmware commands. They normalize and mix the samples, assemble and mix the pools, and dispense the final aliquots. The task tests accurate execution of a large liquid-handling workflow, with scoring based on completed milestones and instrument execution time, excluding time spent reasoning.

Scoring

Reach all six required milestones. Instrument speed does not affect Pass@1.

RESULTS

Normalize 384 samples, combine them into 48 pools, and prepare two aliquots per pool on a simulated Hamilton STARlet.
ModelPass@1Cost ($)Model/API inference cost only, averaged per task attempt. Lab and labor costs are not reported. Repeats are averaged using the same task and family weights as scores.
GPT–6 Astraxhigh100.0%$38.33
Claude Fable 5.1xhigh0.0%$23.93
Claude Opus 5xhigh0.0%$50.73
GPT–5.6 Solxhigh0.0%$37.63
Grok 4.6xhigh0.0%$91.51
Gemini 3.8 Flashhigh0.0%$18.13
0%50%100%

Model trajectories

One example per model.

Run grading

Pass
MeasurementRecorded resultOutcome
Score10.0 %✓ Pass

Pass condition: Reach all six required milestones. Instrument speed does not affect Pass@1.

  • ✓
    Samples normalized

    at 2740.844 s · 785 errors · machine time 129.274 s

  • ✓
    Samples mixed

    at 2740.999 s · 0 errors · machine time 129.297 s

  • ✓
    Pools assembled

    at 2741.173 s · 0 errors · machine time 129.339 s

  • ✓
    Pools mixed

    at 2806.744 s · 0 errors · machine time 144.438 s

  • ✓
    Aliquots made

    at 2807.308 s · 0 errors · machine time 144.548 s

  • ✓
    Finished

    at 2807.485 s · 0 errors · machine time 144.634 s

  • ✓
    Final state

    balances correct; plates seated and tools parked

  • ✓
    Record

    complete

  • ✓
    benchmark environment

    registered

  • ✓
    golden method

    six sealed files match

  • ✓
    agent boundary

    all STARlet work is in interval

  • ✓
    time budget

    2640.74915846996 / 7200 s

  • ✓
    recorded evidence

    complete

  • ✓
    orchestration

    complete

Normalization and poolingGPT–6 Astra xhigh
Run gradingPass
00:0000:01

Model activity and instrument activity

Model transcript

Loading transcript…

Instrument simulation
Computer use