Skip to main content
BuseyBenchcreative model benchmarkSVG models tested
SVG testImage testLeaderboardTimelineMethodology
Browse
SVG testImage testLeaderboardTimelineMethodology
BuseyBenchSame prompt. Same rules. Maximum facial voltage.
Score trendsTimelineMethodologyAboutPrivacyGovernanceRSSAdmin
330 SVG models tested
BuseyBench SVG TestImage Generator Portrait Test

Leaderboard

Judged. Ranked. Busey'd.

One output per model, independent judge scores, and close-score pairwise tie-breaks. The best Gary Busey portraits rise to the top.

Explore the dataDrawing ability over time→
330 models ranked8901 official head-to-head verdictsPairwise tie-break active

Official standings

The rankings

326–330 of 330 results
#3261541 rating ±76
GPT-4o output

GPT-4o

OpenAI / May 13, 2024 / openrouter

Busey likeness0.0
Face coherence0.2
Aesthetics1.2
Prompt adherence0.4

1 judged output · judge spread σ 0.14 · W 3 / L 0 / D 18

0.3overall
#3271500 rating ±76
Granite 4.0 Micro output

Granite 4.0 Micro

IBM / Oct 2, 2025 / openrouter

Busey likeness0.0
Face coherence0.4
Aesthetics1.3
Prompt adherence0.5

1 judged output · judge spread σ 0.05 · W 0 / L 0 / D 21

0.3overall
#3281500 rating ±76
MythoMax 13B output

MythoMax 13B

Gryphe / Jul 2, 2023 / openrouter

Busey likeness0.0
Face coherence0.3
Aesthetics0.9
Prompt adherence0.4

1 judged output · judge spread σ 0.07 · W 0 / L 0 / D 21

0.2overall
#3291404 rating ±82
Llama 3.1 8B Instruct output

Llama 3.1 8B Instruct

Meta / Jul 23, 2024 / openrouter

Busey likeness0.0
Face coherence0.0
Aesthetics0.6
Prompt adherence0.5

1 judged output · judge spread σ 0.05 · W 0 / L 6 / D 12

0.2overall
#3301498 rating ±82
Aion-RP 1.0 (8B) output

Aion-RP 1.0 (8B)

AionLabs / Feb 4, 2025 / openrouter

Busey likeness0.0
Face coherence0.0
Aesthetics0.9
Prompt adherence0.1

1 judged output · judge spread σ 0.06 · W 0 / L 0 / D 18

0.1overall
← PreviousPage 14 of 14Next →
Ranking statusDisplayed score + pairwise tie-break

Overall = likeness 50% + face 20% + aesthetics 15% + prompt 15%. Each model contributes one canonical judged output. The visible overall score, rounded to one decimal, always sets the score band. Pairwise results only order models tied within that displayed band, so a lower displayed score can never leapfrog a higher one.

Read the full methodology →View score trends →
1canonical output judged per model
Closest-scorepairwise matchmaking strategy
busey-v2current scoring rubric