Chose direct inference
The run treated the benchmark as a family of contextual transformations rather than a pure language-model training task.
Run postmortem · Published 2026-08-14
Kimi K2.7 Codexhigh reasoning
Twenty-nine submissions expanded a solver that never became broad enough.
3/ 100
hidden exact
hidden exact
One-hour track · xhigh reasoning · provisional
ARI score scaleexact-match accuracy
3%
0255075100
StatusProvisional
Valid seeds1 / 3
Score rank#18
Current cost$3.77
Points / $0.8
Average time52:13
01 · Verdict
A lot of patching, very little transfer
Kimi K2.7 Code built a compact rule-driven inference system and submitted candidates twenty-nine times. The solver could handle several visible task families, but it never moved beyond 52% public calibration and produced only three hidden exact matches.
One valid autonomous run produced a provisional 3% hidden exact-match score.
#18
Score position3% 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 Kimi K2.7 Code built
The final artifact prioritized explicit solvers over a large trained model.
01
Broad deterministic parser
It implemented records, mappings, arithmetic, sorting, sequences, time operations, text transforms, and odd-one-out logic.
02
Small final package
The selected artifact was only 1.84 MB, leaving size headroom for more coverage.
03
High submission cadence
Twenty-nine candidates show that the run repeatedly patched and tested the solver rather than waiting for a final build.
03 · Trajectory
How the run unfolded
Expanded task coverage
The agent kept adding solver branches and resubmitting, but public accuracy stalled near half.
Finalized below readiness
The hidden exam returned 3%, confirming that the visible parser remained brittle.
04 · Assessment
Good, bad, and unresolved
Correct strategic direction
Direct contextual solving was the same broad mechanism used by the benchmark's stronger systems.
Patches did not create general rules
High submission volume without a strong public score suggests narrow fixes rather than robust abstractions.
Whether the parser could be consolidated
The archive cannot show whether fewer, more general mechanisms would have transferred better than many special cases.
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-0045 | 3% | 52:13 | $3.77 | 1.84 MB | 29 | Valid |
338.2KInput tokens
72.2KOutput tokens
21.8MCache-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 29 candidate submission actions across the published run set. Selected artifacts averaged 1.84 MB compressed.
03
Time use
The runs averaged 52:13 of wall time. 1 run explicitly finalized before the one-hour limit.
Retained totals: 192 agent messages · 29 candidate actions · 1 published run.
06 · Economics
What the result cost
$3.77 per displayed run at current configured rates.
Kimi K2.7 Code was neither a cheap experiment nor a competitive scorer. Most of the spend funded repeated solver iteration that did not generalize.
At current configured API-equivalent rates, the displayed run cost is $3.77 and value is 0.8 score points per dollar.
#19
Value positionValue rank #19 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 Kimi K2.7 Code with the complete ARI Bench field, including score, current cost, value per dollar, effort, and seed status.
View the leaderboard