Benchmark 8 of 18, 10 tasks, asked for 30 s to 32 h

TUA-Bench

Everyday computer chores done entirely from a terminal: editing documents, managing email, and looking up information on the live web.

By agent

Agents on TUA-Bench ranked by the share of their runs within 5% of the time asked, with timing error, where the runs ended and score
RankAgentWithin 5%0.95× to 1.05× of the requestWhere runs ended, from a tenth of the time asked to ten times itTiming erroron this benchmarkScoreout of 100
1 GPT 6.1 SolCodex 55%11 of 20 runs 1.10× 75.110 graded
2 GPT 6 AstraCodex 50%10 of 20 runs 1.13× 92.010 graded
3 GPT 5.6 SolCodex 30%6 of 20 runs 1.42× 66.110 graded
4 Gemini 4 ArgonAntigravity CLI 15%3 of 20 runs 3.03× 68.410 graded
5 Claude Fable 5.1Claude Code 5%1 of 20 runs 3.52× 73.910 graded
6 MiMo V2.6 ProMiMo Code 5%1 of 20 runs 4.42× 50.910 graded
7 MiMo V2.6 Pro UltraSpeedMiMo Code 5%1 of 20 runs 6.57× 67.810 graded
8 Claude Opus 5.5Claude Code 0%0 of 20 runs 2.27× 76.110 graded
9 Muse Wonder 1.0Muse Code 0%0 of 19 runs 11.94× 66.210 graded

Score is the mean of each agent’s graded runs here, out of 100: a v1.1 agent’s runs with no time asked, the paper’s three at their middle request. How it is measured

Every TUA-Bench run

Zoom the chart
View
Typical run, by ask
Short ask
Long ask

Muse Wonder's typical run lasted 0.82× the time asked at the short ask and 0.01× at the long one. Sol 6.1's lasted 1.12× and 1.01×.

Hover a mark to read it, click to pin it here.

Showing 179 runs.

The 10 tasks

The 10 tasks on TUA-Bench, with where each agent's runs ended
TaskAsked forAstraSolFableOpus 5.5ArgonSol 6.1MiMo ProMiMo UltraSpeedMuse Wonder
APA references review1 min to 3.33 hShort 1.10×, long 1.01×Short 1.45×, long 1.03×Short 3.73×, long 0.23×Short 1.10×, long 0.24×Short 13.39×, long 1.18×Short 1.11×, long 1.02×Short 0.77×, long 0.06×Short 0.24×, long 0.05×Short 1.08×, long 0.01×
Daily email report37.5 s to 2.33 hShort 1.15×, long 1.01×Short 1.21×, long 1.03×Short 0.79×, long 0.09×Short 0.78×, long 0.20×Short 8.06×, long 1.04×Short 1.25×, long 1.02×Short 0.35×, long 0.05×Short 0.41×, long 0.03×Short 0.28×, long 0.02×
Author homepage bookmarks37.5 s to 2 hShort 1.13×, long 1.01×Short 1.49×, long 1.02×Short 1.11×, long 0.15×Short 0.77×, long 0.40×Short 6.50×, long 1.13×Short 1.13×, long 1.01×Short 2.02×, long 0.17×Short 1.02×, long 0.05×Short 0.94×, long 0.01×
Name mountain photos37.5 s to 1.67 hShort 1.22×, long 1.02×Short 1.45×, long 1.05×Short 0.89×, long 0.11×Short 1.97×, long 0.59×Short 32.03×, long 1.09×Short 1.11×, long 1.01×Short 0.22×, long 0.08×Short 0.07×, long 0.02×Short 0.76×, long 0.01×
Merge text document30 s to 1.2 hShort 1.44×, long 1.01×Short 3.05×, long 1.06×Short 2.79×, long 0.49×Short 2.02×, long 0.57×Short 7.48×, long 1.25×Short 1.19×, long 1.02×Short 0.94×, long 0.19×Short 0.21×, long 0.14×Short 0.72×, long 0.02×
Place heater for sensors2 min to 10 hShort 1.08×, long 1.01×Short 2.19×, long 1.01×Short 30.00×, long 0.45×Short 1.68×, long 0.39×Short 24.08×, long 1.60×Short 1.06×, long 1.02×Short 4.04×, long 0.38×Short 0.65×, long 0.15×Short 2.38×, long no run
Optimize cold plate6 min to 32 hShort 1.08×, long 1.01×Short 1.65×, long 1.01×Short 2.96×, long 0.36×Short 0.33×, long 0.36×Short 2.54×, long 0.83×Short 1.03×, long 1.00×Short 0.69×, long 0.04×Short 0.44×, long 0.07×Short 0.06×, long 0.01×
Extract gym auditorium1.25 min to 5 hShort 1.29×, long 1.02×Short 5.68×, long 1.08×Short 17.23×, long 1.11×Short 3.21×, long 0.28×Short 4.21×, long 1.01×Short 1.22×, long 1.01×Short 0.96×, long 0.11×Short 0.74×, long 0.04×Short 0.14×, long 0.01×
EPW parquet check1.25 min to 1.67 hShort 2.47×, long 1.02×Short 2.88×, long 1.08×Short 4.86×, long 0.49×Short 1.75×, long 0.50×Short 4.43×, long 0.48×Short 2.15×, long 1.02×Short 2.89×, long 0.16×Short 1.58×, long 0.10×Short 1.46×, long 0.11×
Extract presenter photos2 min to 4.33 hShort 1.14×, long 1.02×Short 1.11×, long 1.04×Short 1.05×, long 0.13×Short 0.61×, long 0.21×Short 2.74×, long 1.03×Short 1.07×, long 1.01×Short 2.16×, long 0.11×Short 0.57×, long 0.04×Short 0.89×, long 0.00×

Each mark is one run, at its short or its long ask, placed from 0.1× to 10× of the time asked; the pale band is within 5% of it.

About TUA-Bench: the prompt, the grader, the source

Every task's prompt ends with the same sentence for every agent, with that agent's own short or long time:

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.

Every run is graded by TUA-Bench's own grader, not a scorer AgentTime invented. The pass rule: Full completion: a reward of 1 on the final state.

Tasks
10
Asked for
30 s to 32 h
Runs here
179
9 agents
Graded as
Pass or fail: Full completion: a reward of 1 on the final state.
Data release
2 Nov 2026
17:20 UTC

Read before quoting. 10 runs here were stopped by AgentTime, so their time is a lower bound. Details

Data and citation

Upstream: Chen et al., 2026