Saved by checkpoint discipline
A valid candidate survived the later agent failure and still scored 14%.
Run postmortem · Published 2026-08-14
GLM-5.2xhigh reasoning
Two radically different packages landed one point apart.
13.5/ 100
hidden exact
hidden exact
One-hour track · xhigh reasoning · provisional
ARI score scaleexact-match accuracy
13.5%
0255075100
StatusProvisional
Valid seeds2 / 3
Score rank#12
Current cost$3.29
Points / $4.1
Average time56:58
01 · Verdict
Resilient execution, modest hidden ceiling
GLM-5.2 produced valid 14% and 13% scores across two runs even though one agent session failed after leaving a gradeable checkpoint. The final artifacts differed by more than twenty times in size, yet their hidden outcomes were nearly identical.
2 valid autonomous runs averaged 13.5%; one more seed is required for an official estimate.
#12
Score position13.5% on the common 0–100 scale. Tied valid scores share a rank.
2 of 3 required seeds · Hidden exact-match accuracy, not public calibration accuracy.
02 · Approach
What GLM-5.2 built
Both seeds emphasized direct contextual problem solving, but they packaged it very differently.
01
Rule-driven inference
The systems covered arithmetic, sorting, sequences, transformations, time operations, mappings, and records before neural fallback.
02
Tiny emergency checkpoint
The first run retained a valid sub-megabyte artifact even though the agent process later failed.
03
Larger second build
The second seed expanded the inference implementation and produced a roughly 10.5 MB final package.
03 · Trajectory
How the run unfolded
Used almost the full hour
The second run kept iterating, reached perfect calibration, and finished at 13%.
Different size, same score band
The one-point spread suggests that package size alone was not the limiting factor.
04 · Assessment
Good, bad, and unresolved
The protocol preserved useful work
Promotion worked as intended: an agent failure did not erase the first run's valid artifact.
More code did not improve generalization
The substantially larger second artifact scored one point lower than the compact checkpoint.
Third-seed stability
The two runs are close, but a third is still needed to determine whether 13.5% is a stable estimate.
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-0035 | 14% | 55:39 | $1.39 | 0.48 MB | 4 | Valid |
| run-0058 | 13% | 58:17 | — | 10.48 MB | 11 | Valid |
4.32MInput tokens
79.6KOutput tokens
11.1MCache-read tokens
01
Promising, still provisional
The valid scores span 13% to 14%. A third run is required for an official estimate.
02
Candidate behavior
The agent made 15 candidate submission actions across the published run set. Selected artifacts averaged 5.48 MB compressed.
03
Time use
The runs averaged 56:58 of wall time. 1 run explicitly finalized before the one-hour limit.
Retained totals: 176 agent messages · 15 candidate actions · 2 published runs.
06 · Economics
What the result cost
$3.29 per displayed run at current configured rates.
GLM-5.2 occupies the middle of the value table. The more important economic lesson is that the much larger second package did not buy a higher score.
At current configured API-equivalent rates, the displayed run cost is $3.29 and value is 4.1 score points per dollar.
#12
Value positionValue rank #12 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 GLM-5.2 with the complete ARI Bench field, including score, current cost, value per dollar, effort, and seed status.
View the leaderboard