Entered the one-hour track at xhigh
The model began from the same public materials, hidden-exam boundary, and artifact constraints used across the leaderboard.
Run postmortem · Published 2026-08-20
Qwen3.8-27B Aggressive FastMTP Q3_K_Pxhigh reasoning
A Q3_K_P HauhauCS derivative with FastMTP reached 19% on one RTX 5090—five points above the earlier local Qwen run.
19/ 100
hidden exact
hidden exact
One-hour track · xhigh reasoning · provisional
ARI score scaleexact-match accuracy
19%
0255075100
StatusProvisional
Valid seeds1 / 3
Score rank#5
Current cost$0.00
Points / $—
Average time55:49
01 · Verdict
A materially stronger, distinct single-GPU Qwen result
This is not a rerun of the previously published 14% Qwen3.8-27B entry. It used the HauhauCS Aggressive GGUF derivative at Q3_K_P with a FastMTP sidecar at depth 3, xhigh agent effort, and a 200K-class context configuration on one RTX 5090. The selected artifact passed validation and answered 19 of 100 hidden questions exactly. With one seed, the result remains provisional.
One valid autonomous run produced a provisional 19% hidden exact-match score.
#5
Score position19% 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 Qwen3.8-27B Aggressive FastMTP Q3_K_P built
The run combined a quantized local agent model with speculative multi-token prediction while preserving the standard one-hour ARI Bench artifact workflow.
01
Distinct HauhauCS checkpoint
The agent used the Aggressive Qwen3.8-27B GGUF derivative rather than the NVFP4 checkpoint behind the earlier 14% row.
02
FastMTP sidecar at depth 3
Speculative multi-token prediction accelerated generation while the main model retained final-token verification.
03
Single RTX 5090 deployment
Inference remained local and unmetered, with xhigh effort and a large-context configuration tuned to fit one consumer GPU.
03 · Trajectory
How the run unfolded
Maintained an iterative candidate loop
Five candidate actions were retained, and the selected package reached 25 of 25 on public calibration.
Finalized a valid 19% artifact
The final 11.94 MB compressed package passed grading and answered 19 of 100 hidden items exactly.
04 · Assessment
Good, bad, and unresolved
Near-frontier one-seed local result
The run matched the 19% score of a strong API-served entry while using one locally owned RTX 5090 and no metered API calls.
Throughput weakened at the longest active context
Generation usually exceeded 100 tokens per second at shorter and moderate context, but dropped below that target as the active context approached roughly 131K tokens.
Repeatability and component attribution
Two more seeds are needed, and one aggregate score cannot isolate how much of the gain came from the checkpoint, quantization, sidecar, context, or run variance.
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-0100 | 19% | 55:49 | $0.00 | 11.94 MB | 5 | Valid |
298.8KInput tokens
187.8KOutput tokens
15MCache-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 5 candidate submission actions across the published run set. Selected artifacts averaged 11.94 MB compressed.
03
Time use
The runs averaged 55:49 of wall time. 1 run explicitly finalized before the one-hour limit.
Retained totals: 156 agent messages · 5 candidate actions · 1 published run.
06 · Economics
What the result cost
$0.00 per displayed run at current configured rates.
The run incurred no metered API charge because inference was local. Hardware ownership and electricity are excluded, so points-per-dollar is not reported.
At current configured API-equivalent rates, the displayed run cost is $0.00 and value is — score points per dollar.
—
Value positionValue rank unavailable. Higher score points per dollar indicate better value.
Current configured API-equivalent cost / run · pricing snapshot 2026-08-20.
07 · Context
Nearby results
Adjacent rows provide score context without treating small one-seed differences as settled model rankings.
Compare Qwen3.8-27B Aggressive FastMTP Q3_K_P with the complete ARI Bench field, including score, current cost, value per dollar, effort, and seed status.
View the leaderboard