Benchmark 10 of 18, 12 tasks, asked for 30 s to 5 h

BrowseComp

Hard-to-find facts on the live web: each question has a short answer that takes many searches to pin down.

By agent

Agents on BrowseComp 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 63%15 of 24 runs 1.08× 83.312 graded
2 Gemini 4 ArgonAntigravity CLI 33%8 of 24 runs 1.33× 75.012 graded
3 Muse Wonder 1.0Muse Code 21%5 of 24 runs 2.77× 75.012 graded
4 Claude Opus 5.5Claude Code 17%4 of 24 runs 1.75× 75.012 graded
5 MiMo V2.6 ProMiMo Code 0%0 of 24 runs 3.90× 66.712 graded
6 MiMo V2.6 Pro UltraSpeedMiMo Code 0%0 of 24 runs 6.60× 58.312 graded

Score is the mean of each agent’s graded runs here, out of 100: its runs with no time asked. How it is measured

Every BrowseComp run

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

MiMo UltraSpeed's typical run lasted 0.51× the time asked at the short ask and 0.04× at the long one. Sol 6.1's lasted 1.11× and 1.02×.

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

Showing 144 runs.

The 12 tasks

The 12 tasks on BrowseComp, with where each agent's runs ended
TaskAsked forOpus 5.5ArgonSol 6.1MiMo ProMiMo UltraSpeedMuse Wonder
Question 13.5 min to 5 hShort 6.64×, long 0.58×Short 1.18×, long 1.00×Short 1.03×, long 1.01×Short 3.31×, long 0.31×Short 0.52×, long 0.06×Short 1.96×, long 0.95×
Question 245 s to 3.67 hShort 1.43×, long 1.11×Short 1.30×, long 1.03×Short 1.44×, long 1.04×Short 0.94×, long 0.03×Short 0.48×, long 0.02×Short 0.95×, long 0.06×
Question 31.5 min to 4.67 hShort 11.63×, long 0.87×Short 1.14×, long 1.01×Short 1.03×, long 1.01×Short 1.89×, long 0.06×Short 1.28×, long 0.09×Short 2.70×, long 0.22×
Question 42 min to 2.8 hShort 3.52×, long 1.00×Short 3.93×, long 1.01×Short 1.06×, long 1.04×Short 3.83×, long 0.21×Short 0.53×, long 0.04×Short 1.19×, long 0.09×
Question 51.25 min to 2.33 hShort 1.07×, long 0.56×Short 1.19×, long 1.01×Short 1.03×, long 1.01×Short 3.39×, long 0.40×Short 0.40×, long 0.13×Short 1.06×, long 1.02×
Question 645 s to 2.2 hShort 1.33×, long 1.04×Short 1.22×, long 1.09×Short 1.16×, long 1.02×Short 2.19×, long 0.22×Short 0.27×, long 0.02×Short 1.18×, long 0.05×
Question 730 s to 2.33 hShort 0.88×, long 1.05×Short 2.77×, long 1.06×Short 1.32×, long 1.01×Short 3.30×, long 0.33×Short 1.64×, long 0.03×Short 0.56×, long 0.02×
Question 852.5 s to 2.8 hShort 1.48×, long 0.82×Short 1.51×, long 1.05×Short 1.16×, long 1.01×Short 2.55×, long 0.22×Short 0.50×, long 0.08×Short 1.69×, long 0.06×
Question 945 s to 2.33 hShort 2.64×, long 1.03×Short 1.90×, long 1.10×Short 1.11×, long 1.02×Short 7.68×, long 0.19×Short 0.73×, long 0.02×Short 1.02×, long 1.01×
Question 103 min to 4 hShort 5.81×, long 1.00×Short 3.20×, long 1.01×Short 1.08×, long 1.01×Short 2.46×, long 0.50×Short 0.63×, long 0.13×Short 1.55×, long 1.03×
Question 1152.5 s to 4.33 hShort 1.49×, long 0.67×Short 1.34×, long 1.01×Short 1.19×, long 1.03×Short 3.11×, long 0.11×Short 0.20×, long 0.03×Short 1.20×, long 0.20×
Question 121 min to 3 hShort 3.27×, long 1.19×Short 1.47×, long 1.19×Short 1.10×, long 1.02×Short 4.05×, long 0.33×Short 0.49×, long 0.05×Short 0.68×, long 0.13×

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 BrowseComp: 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 BrowseComp's own grader, not a scorer AgentTime invented. It reports a score: whether the final answer matches the reference answer, 100 or 0.

Tasks
12
Asked for
30 s to 5 h
Runs here
144
6 agents
Graded as
whether the final answer matches the reference answer, 100 or 0
Data release
2 Nov 2026
17:20 UTC

Data and citation

Upstream: Wei et al., 2025