TL Task Lens

Three real computer-use tests

Do the work.
Prove it worked.

These agents must operate specialist software—not merely write an answer.

GPT-6 AstraOpenAI-reported result
59.3%Agents’ Last Exam

The best result in the published comparison.

Claude Opus 5 scored 55.5%. GPT-5.6 Sol scored 53.6%.
At the top settings, Astra used about 65% fewer output tokens than Opus 5.Source ↗
01 Manufacturing

The question, in plain English

Turn this metal block into the right part.

Open industrial machining software. Plan every cut, avoid crashes, simulate the job and save the finished project.

10,000surface checks
0.3 mmcritical tolerance
0crashes allowed
Reference → expected result
Reference view of the metal part to be machinedFinished machining project in Autodesk PowerMill
Go deeperHover or tap any point
AnswerA working project

Safe roughing and finishing toolpaths, with a simulated final part that matches the target surface.

Required evidenceagent_sim.stl
Open the source task ↗
02 Engineering

The question, in plain English

Will molten plastic fill this mould properly?

Set the manufacturing conditions. Run the full moulding simulation. Extract the exact results.

4simulation stages
1–2 hrrun time
9results to report
Setup → completed analysis
Moldex3D software showing the supplied mould projectMoldex3D displaying a completed mould-flow analysis
Go deeperHover or tap any point
Answer9 verified readings

A completed Fill → Pack → Cool → Warp analysis, plus its key pressure, temperature, time and deformation values.

Required fileresults.json
Open the source task ↗
03 Game design

The question, in plain English

Rebuild the first three floors of this old game.

Study the original Flash game. Recreate its maps, monsters and battles in RPG Maker XP. Then play it.

3playable floors
2different programs
3proof screenshots
Original → editor → rebuilt game
The original tower game running through a Flash emulatorThe recreated tower level inside RPG Maker XPReference view of a completed game floor
Go deeperHover or tap any point
AnswerA playable remake

Three recognisable floors with matching layouts, a working player, monsters, combat and stairs.

Required proof3 screenshots
Open the source task ↗