Ranked 6 of 9 on the combined score
Gemini 4 Argon
Antigravity CLI, high reasoning. 400 timed runs on 200 tasks from 18 benchmarks.
Duration following
Does it work for as long as it is asked? Each task was asked for a short and a long duration: a quarter and four times its own forecast of each task (on PaperBench and PostTrainBench, of its runtime with no time asked).
| Ask | Timing error | Where runs ended |
|---|---|---|
| Every timed run400 runs | 2.62× | |
| Short ask200 runs | 3.69× | |
| Long ask200 runs | 1.87× |
It ran long on short asks (median 4.15× at the short ask) and was within 5% of the time asked on long ones (median 1.01× at the long ask). See every run
Every Argon run
Each row is one run, with a mark on the chart. Hover a row to find its mark; click a mark to find its row.
400 runs
| Loading 400 runs. | |||
No runs match these filters.
Forecasting
Before each task, it forecast how long the task would take. Each forecast is set against its own runtime on that task: its run with no time asked.
- Typical miss
- 2.59×
- Geometric mean of the factor between forecast and runtime
- Most forecasts
- 79% too low
- On 200 tasks
- Median forecast
- 12 min
- Median runtime
- 25 min
Retrospection
After its runs had finished, it was asked how long they took, with less time information at each step. Each figure is its typical miss against the real runtime, two answers per condition.
- Oracle
- 1.00×
- Native
- 1.75×
- Context-only
- 2.47×
- Replay
- 2.06×
- Scrubbed
- 3.73×
Oracle adds an elapsed-time tool. Native and Context-only continue the finished session, with tools on or off. Replay sends a rebuilt transcript to the model's API, and Scrubbed does the same without timestamps or other time cues. On 23 tasks.
Across the 18 benchmarks
Timing on its timed runs, and how good the work was. Score is the mean of its graded runs out of 100, from its runs with no time asked; the benchmark score is the mean of these, each benchmark counting the same.
| OSWorld36 runs | 4.04× | 8% | 66.418 graded |
| AppWorld28 runs | 3.95× | 7% | 57.114 graded |
| METR public tasks4 runs | 3.59×(few runs) | 0% | Too few graded |
| PPTArena20 runs | 3.40× | 10% | 43.010 graded |
| Humanity's Last Exam40 runs | 3.19× | 5% | 65.020 graded |
| GPQA Diamond24 runs | 3.13× | 0% | 100.012 graded |
| TUA-Bench20 runs | 3.03× | 15% | 68.410 graded |
| Terminal-Bench 4.032 runs | 2.92× | 6% | 75.016 graded |
| ProgramBench36 runs | 2.91× | 6% | 59.118 graded |
| CORE-Bench24 runs | 2.87× | 8% | 75.012 graded |
| DeepSWE28 runs | 2.77× | 4% | 92.914 graded |
| PostTrainBench8 runs | 2.26×(few runs) | 25% | Not scored |
| PaperBench8 runs | 2.22×(few runs) | 0% | Not scored |
| Agents' Last Exam24 runs | 2.16× | 8% | 72.212 graded |
| YC-Bench8 runs | 1.98×(few runs) | 0% | Own units |
| Sakana ALE-Bench20 runs | 1.84× | 25% | Own units |
| AssistantBench16 runs | 1.65× | 38% | 40.78 graded |
| BrowseComp24 runs | 1.33× | 33% | 75.012 graded |
No runs match these filters.
Highest timing error first. ALE-Bench and YC-Bench keep their own units and stay out of the score, as do PaperBench and PostTrainBench.
Read before quoting. 11 runs were stopped by AgentTime, so their time is a lower bound. Details