GUI-World · matched GPT-6 Luna reader

Näky structured output vs 20 screenshots

Absolute task quality, paired differences, exact reader input tokens, estimated API cost, latency, and predefined subgroup results from the authenticated three-trial benchmark.

Matched reader policy: GPT-6 Luna, low reasoning, 16,384 maximum output tokens, high image detail, no storage, and high service tier. The screenshot arm uses 20 frames at a 768-pixel long edge sampled nearest each equal-presentation-duration-bin midpoint.

Näky accuracy82.61%2,865 / 3,468 tasks
20-screenshot accuracy88.21%3,059 / 3,468 tasks
Paired difference-5.59 pp95% CI -6.82 pp to -4.38 pp
Input-token saving50.91%exact unique body counts

Quality

ArmCorrectTask accuracyScenario macroPooled-call accuracyInvalid tasks
Näky structured output2,865 / 3,46882.61%81.81%82.36%38
20 screenshots3,059 / 3,46888.21%87.77%87.75%10

Paired outcomes: 131 Näky wins, 325 losses, 2,994 ties, 18 unresolved. Confidence intervals use the score’s predefined scenario-stratified media-cluster bootstrap.

Tokens, cost, and latency

ArmUnique input tokens3-trial planned tokensObserved input tokensEstimated costp50 msp95 ms
Näky structured output12,413,47637,240,42837,235,285$5.26479545,326
20 screenshots25,289,40875,868,22475,868,224$9.84101,9384,716

Cost is the score’s cache-aware estimate under its frozen price table, not a provider invoice. Latency covers authenticated responses in the complete evaluable cohort.

Product processing and evidence size

MetricValue
Evaluable recordings1,819
Normalized video27,124.27 s
ScreenEvents1,378,711
Record wall-time sum31,011.66 s
Record wall-time mean17.05 s
Record wall-time p5011.01 s
Record wall-time p9552.75 s
Video seconds per record wall second0.875
Peak record RSS679,280 KiB (663.36 MiB)
Näky state text13,639,728 (13.01 MiB)
Screenshot PNG evidence5,941,847,499 (5.53 GiB)

Record wall-time values are per-record compute work, not parallel batch makespan. Evidence sizes cover the complete 1,819-record evaluable corpus and are separate from reasoning-model input tokens.

Complete-case sensitivity

exclude the task pair affected by ambiguous transport. Excluded missing tasks: 1.

ArmAccuracyCorrectTasks
Näky structured output82.61%28643467
20 screenshots88.20%30583467

Complete-case paired delta: -5.60 pp.

Interpretation

terminal-tolerant sensitivity with one authenticated ambiguous transport; no noninferiority or equivalence claim

Reproducibility scope

The 1,819-record equivalence run executed the authenticated unstripped binary. The release executable is its twice-reproduced GNU strip --strip-all derivative; the stripped file was not rerun across the full corpus.

Strict evaluated equivalence uses an authenticated mixed-host composition: 1,815 records from the base run and four AVX2/FMA/no-AVX-512 replacements. The base-to-replacement diagnostic differed in 4 of 1,819 event streams and 2 of 1,819 screen bodies. The receipt does not bind the base host ISA, so cross-CPU byte invariance is not claimed.

Binary provenance

RoleSHA-256BytesExecution scope
Scored product binaryf3d5a318ff3571c1f1a1aafb22d21d9257722eef3d16522c7c2cc7c8901bf9e2—Three-trial reader benchmark
Public equivalence execution binarya80e32664cd467fda606d3a6b1256ad0d26a5a9ad0e166353631fb18c0fc88f112,912,9521,819-record output-equivalence run
Stripped release binaryc791889fa5bc3f85a8398dddc0901ff3cf4f9bb84777fe48bf7e1f3c58c33b4a9,457,304Deterministic derivative; full corpus not rerun

Scenario

GroupTasksMediaNäky20 screenshotsDelta95% CI
IOS33016482.73%90.61%-7.88 pp-12.20 pp to -3.64 pp
XR28814475.35%83.68%-8.33 pp-12.50 pp to -4.17 pp
android118459282.52%88.09%-5.57 pp-7.69 pp to -3.38 pp
multi26613380.08%83.83%-3.76 pp-8.65 pp to +1.13 pp
software93446183.73%88.12%-4.39 pp-6.70 pp to -2.05 pp
website46623386.48%92.27%-5.79 pp-8.58 pp to -3.00 pp

Platform

GroupTasksMediaNäky20 screenshotsDelta95% CI
desktop140069484.64%89.50%-4.86 pp-6.64 pp to -3.07 pp
mobile151475682.56%88.64%-6.08 pp-8.05 pp to -4.16 pp
multi26613380.08%83.83%-3.76 pp-8.27 pp to +0.75 pp
xr28814475.35%83.68%-8.33 pp-12.85 pp to -4.17 pp

Task type

GroupTasksMediaNäky20 screenshotsDelta95% CI
mcqa1735172782.82%89.97%-7.15 pp-8.88 pp to -5.43 pp
reasoning1733172582.40%86.44%-4.04 pp-5.65 pp to -2.42 pp

Duration

GroupTasksMediaNäky20 screenshotsDelta95% CI
lt-152198109681.57%88.22%-6.64 pp-8.16 pp to -5.14 pp
15-to-lt-3097848783.33%88.34%-5.01 pp-7.35 pp to -2.65 pp
30-to-lt-6026212987.79%87.79%+0.00 pp-3.41 pp to +3.44 pp
60-to-lt-120281489.29%85.71%+3.57 pp-10.71 pp to +17.86 pp
gte-12021100.00%100.00%+0.00 pp+0.00 pp to +0.00 pp

Resolution

GroupTasksMediaNäky20 screenshotsDelta95% CI
lte-720120060082.67%88.17%-5.50 pp-7.67 pp to -3.42 pp
721-to-1080201085.00%90.00%-5.00 pp-15.00 pp to +0.00 pp
gt-10802248111782.56%88.21%-5.65 pp-7.15 pp to -4.17 pp

Orientation

GroupTasksMediaNäky20 screenshotsDelta95% CI
portrait139869882.62%88.56%-5.94 pp-7.95 pp to -3.94 pp
landscape2070102982.61%87.97%-5.36 pp-6.91 pp to -3.82 pp
square00————