P/RProofrunIndependent model tests / 010
Results brief 010 / Interface design / Published August 20, 2026

glm-5.3 vs gemini-3.7-flash: Orbit Radio

A tactile sci-fi tuner that tests hierarchy, motion, and input handling.

Bounded resultz-ai/glm-5.3 won this frozen run.Structured from the locked human evaluation. One run; no broader ranking is implied.
3 min readAugust 20, 2026Interface designproofrun-v0.7
Verdictz-ai/glm-5.3
Confidencelow
Blind score46:34
Margin+12
01 / Publication evidence

The result,
shown before explained.

A
Contestant Az-ai/glm-5.3
Winner
Publication evidenceContestant A · Orbit Radio
Contestant A populated Orbit Radio capture
1280 × 720 original capture

Next was activated twice and Previous once from a fresh Orbit Radio load, then the resulting signal state was captured after the standardized stabilization window.

Contestant A · preserved capture
Score46
Wall time1:28
HTML start0:05
Cost$0.0222
B
Contestant Bgoogle/gemini-3.7-flash
Runner-up
Publication evidenceContestant B · Orbit Radio
Contestant B populated Orbit Radio capture
1280 × 720 original capture

Next was activated twice and Previous once from a fresh Orbit Radio load, then the resulting signal state was captured after the standardized stabilization window.

Contestant B · preserved capture
Score34
Wall time0:15
HTML start0:01
Cost$0.0256
“This one was very close but Contestant A takes the win here because of the slightly nicer UI, despite the drag and scroll text being slightly behind the previous button Contestant As central dial is smooth and has no issues while dragging it while Contestant Bs central dial seems to jump around. Contestant As visualizations are also better in my opinion.”Locked before model reveal
02 / Frozen evaluation

Requirements first.
Weighted quality second.

Frozen requirements6 judgments per artifact
RequirementAB
The central dial responds to turning or dragging.MetMet
Five distinct fictional stations are available.MetMet
Station name, frequency, color, and visualization change together.MetMet
Previous and next controls work consistently.MetMet
The current station is visually unmistakable.MetMet
The single-file instrument remains responsive and works offline.MetMet
Weighted criteriaA 46 / B 34

Visual design · 30%

Hierarchy, composition, typography, color, polish, and coherent visual judgment.

A / 105
B / 104

Interaction & usability · 30%

Discoverability, responsiveness, feedback, control quality, and interaction feel.

A / 105
B / 104

Technical robustness · 30%

Runtime correctness, edge-case handling, performance, responsiveness, and code reliability.

A / 105
B / 104

Originality · 10%

Useful creative choices that go beyond a generic first solution without harming clarity.

A / 106
B / 105
03 / Run integrity

The claim, with its
conditions attached.

Contestant A1:28$0.0222 reported cost
Contestant B0:15$0.0256 reported cost
Protocolv0.7curated:orbit-radio:v3
Immutable attempts and publication recordInspect provenance

This Results Brief is rendered from a sanitized, allowlisted record rather than AI-authored article prose. Credentials, raw responses, generated source, executable artifacts, and private reviewer notes remain in the private Studio.

Run recordPR-MT1JXMC5-7F
Outcomehead to head
Test lineagev3 · curated:orbit-radio:v3
Identityrevealed
Evaluationlocked
Record hashsha256:a18cc0cd707db557113d4d181a8a036b9a7e3ad956909b06a742c3f606a2afb8
Immutable attempt ledger

A #1 original: completed submission (normal, Z.AI) · B #1 original: completed submission (normal, Google AI Studio)

Download the hashed publication record ↓