Built the direct solver first
The run focused immediately on task-aware inference rather than lengthy model training.
Run postmortem · Published 2026-08-14
Gemini 3.6 Flashxhigh reasoning
The shortest generated run produced a 14% score with a 3.2 MB artifact.
14/ 100
hidden exact
hidden exact
One-hour track · xhigh reasoning · provisional
ARI score scaleexact-match accuracy
14%
0255075100
StatusProvisional
Valid seeds1 / 3
Score rank#11
Current cost$1.18
Points / $11.9
Average time6:33
01 · Verdict
A remarkably efficient checkpoint, not a full-hour search
Gemini 3.6 Flash assembled a small deterministic and n-gram hybrid, reached perfect public calibration, and finalized in six and a half minutes. Its 14% hidden score is strong relative to both time and artifact size, but it leaves the largest unused exploration budget of any successful generated page.
One valid autonomous run produced a provisional 14% hidden exact-match score.
#11
Score position14% on the common 0–100 scale. Tied valid scores share a rank.
1 of 3 required seeds · Hidden exact-match accuracy, not public calibration accuracy.
02 · Approach
What Gemini 3.6 Flash built
The final package emphasized direct inference and lightweight statistical fallback over a large trained network.
01
Rule-based problem solving
The solver covered arithmetic, structured records, mappings, sequences, and contextual extraction.
02
N-gram fallback
When a direct answer was unavailable, the system used local byte statistics rather than a large neural checkpoint.
03
Small final artifact
The selected package was only 3.20 MB and required seven candidate submissions.
03 · Trajectory
How the run unfolded
Saturated calibration
Every visible item passed, removing the public set as a useful selector.
Finalized fifty-three minutes early
The hidden result landed at 14%, well above several full-hour attempts.
04 · Assessment
Good, bad, and unresolved
Exceptional execution efficiency
The run converted very little time and a small package into a competitive upper-midfield score.
No attempt to spend the remaining budget
The agent did not use the hour to diversify tests after public calibration saturated.
Whether 14% was a floor or ceiling
With only one seed and six minutes of work, the result says little about what a patient 3.6 Flash run could do.
05 · Evidence
The retained run record
Every archived run selected for this exact model-and-effort row is shown below. The narrative uses aggregate run evidence and final artifact structure; it does not expose hidden questions, answers, or raw private transcripts.
Swipe horizontally to see all run columns.
| Run | Score | Time | Cost at run | Artifact | Candidates | Grade |
|---|---|---|---|---|---|---|
| run-0072 | 14% | 6:33 | $1.18 | 3.20 MB | 7 | Valid |
343.4KInput tokens
38.2KOutput tokens
2.5MCache-read tokens
01
One run, one outcome
Autonomous runs are stochastic. Two more valid seeds are required before this can be treated as an official estimate.
02
Candidate behavior
The agent made 7 candidate submission actions across the published run set. Selected artifacts averaged 3.20 MB compressed.
03
Time use
The runs averaged 6:33 of wall time. 1 run explicitly finalized before the one-hour limit.
Retained totals: 57 agent messages · 7 candidate actions · 1 published run.
06 · Economics
What the result cost
$1.18 per displayed run at current configured rates.
Gemini 3.6 Flash is one of the better value results even without repricing. Its strongest advantage was operational: it achieved 14% with minimal wall time and a small artifact.
At current configured API-equivalent rates, the displayed run cost is $1.18 and value is 11.9 score points per dollar.
#5
Value positionValue rank #5 of 22 valid priced entries. Higher score points per dollar indicate better value.
Current configured API-equivalent cost / run · pricing snapshot retained archive basis.
07 · Context
Nearby results
Adjacent rows provide score context without treating small one-seed differences as settled model rankings.
Compare Gemini 3.6 Flash with the complete ARI Bench field, including score, current cost, value per dollar, effort, and seed status.
View the leaderboard