Community preference data

Blind AI image voting results

Candidate identities are hidden during voting. Arena Elo adjusts for opponent strength without depending on vote order. Ties and “both miss” remain separate from decisive preferences.

Ballots
1832
Voted matchups
461
Qualified matchups
0
Vote in the Arena →Inspect source tests →
Model-level signal

Arena Elo ratings

Scores are produced by an order-independent Bradley-Terry model and centered at 1,000. The chart adjusts for opponent strength and shows a 95% uncertainty range. A numbered rank requires at least 50 decisive votes across 5 prompts and 3 opponents.

  1. GPT Image 2 Medium110 votes · 7 prompts · 8 opponents
    1,111±69
  2. Seedream V5 Pro656 votes · 91 prompts · 8 opponents
    1,092±46
  3. Seedream V5 Lite660 votes · 92 prompts · 8 opponents
    1,057±46
  4. Nano Banana 2203 votes · 27 prompts · 8 opponents
    1,045±55
  5. Seedream 4.5586 votes · 80 prompts · 7 opponents
    1,007±48
  6. Grok Imagine106 votes · 12 prompts · 6 opponents
    996±67
  7. Nano Banana 2 Lite119 votes · 16 prompts · 7 opponents
    974±75
  8. Grok88 votes · 5 prompts · 7 opponents
    949±77
  9. Nano Banana120 votes · 16 prompts · 7 opponents
    917±76
ProvisionalIdeogram 4: 1005 Elo · 42 votes · 5 prompts · 2 opponentsKrea 2 Medium: 846 Elo · 42 votes · 5 prompts · 2 opponents
Matchup results

Results by matchup

21/25 ballots · early · Intent match

Realistic Bathroom Selfie Casual Candid vs Girl In Beige Armchair 2012 Style Selfie

76.2% vs 23.8% among 21 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
20/25 ballots · early · Intent match

Bathroom Mirror Selfie vs Couple Kissing In Dimly Lit Room

70% vs 30% among 20 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
20/25 ballots · early · Intent match

Couple Kissing Warm Soft Lighting Wooden Shelf Framed Pictures vs Couple Kissing In Dimly Lit Room

78.9% vs 21.1% among 19 decisive preferences. 1 tie; 0 both-miss ballots.

Open result →
20/25 ballots · early · Intent match

Blue Shirt Headshot Neutral Background vs Professional Female Studio Portrait

42.1% vs 57.9% among 19 decisive preferences. 1 tie; 0 both-miss ballots.

Open result →
20/25 ballots · early · Intent match

Beach Bikini Woman Wooden Structure Wide Angle Cinematic vs Beach In Swimsuit Lying Prone

95% vs 5% among 20 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
10/25 ballots · early · Prompt vs mannequin

Mannequin vs Prompt

50% vs 50% among 10 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
9/25 ballots · early · Seedream portrait archive

Seedream V5 Lite vs Seedream 4.5

88.9% vs 11.1% among 9 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
8/25 ballots · early · Controlled portrait test

GPT Image 2 Medium vs Grok

87.5% vs 12.5% among 8 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
8/25 ballots · early · Controlled portrait test

GPT Image 2 Medium vs Seedream V5 Lite

71.4% vs 28.6% among 7 decisive preferences. 1 tie; 0 both-miss ballots.

Open result →
8/25 ballots · early · Seedream portrait archive

Seedream V5 Lite vs Seedream 4.5

42.9% vs 57.1% among 7 decisive preferences. 1 tie; 0 both-miss ballots.

Open result →
8/25 ballots · early · Seedream portrait archive

Seedream V5 Pro vs Seedream 4.5

100% vs 0% among 7 decisive preferences. 1 tie; 0 both-miss ballots.

Open result →
8/25 ballots · early · Seedream portrait archive

Seedream 4.5 vs Seedream V5 Pro

0% vs 100% among 8 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
8/25 ballots · early · Prompt vs mannequin

Mannequin vs Prompt

50% vs 50% among 8 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
8/25 ballots · early · Prompt vs mannequin

Mannequin vs Prompt

62.5% vs 37.5% among 8 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
8/25 ballots · early · Prompt vs mannequin

Mannequin vs Prompt

50% vs 50% among 8 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
8/25 ballots · early · Prompt vs mannequin

Mannequin vs Prompt

50% vs 50% among 8 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →
7/25 ballots · early · Controlled portrait test

Seedream V5 Lite vs Seedream 4.5

50% vs 50% among 6 decisive preferences. 0 ties; 1 both-miss ballot.

Open result →
7/25 ballots · early · Seedream portrait archive

Seedream V5 Pro vs Seedream V5 Lite

42.9% vs 57.1% among 7 decisive preferences. 0 ties; 0 both-miss ballots.

Open result →