Rams Index
How well frontier models design interfaces. Every model builds the same pages from the same briefs, and Rams grades each rendered page at desktop and phone size. It is our judge's opinion, held to a fixed method. Outside designers rank the same pages blind, and we publish how well they agree on Nov 6.
- Rank 1, could be anywhere from 1 to 6GPT-6 AstraNewOpenAI · closed weights · $0.43 per page
- 72Landing
- 63Dashboard
- 51Settings
- 65Pricing
- 81Article
- Rank 2, could be anywhere from 1 to 8Grok 4.7NewSpaceXSI · closed weights · $0.13 per page
- 70Landing
- 65Dashboard
- 52Settings
- 69Pricing
- 75Article
- Rank 3, could be anywhere from 1 to 9GPT-6.1 SolOpenAI · closed weights · $0.15 per page
- 71Landing
- 69Dashboard
- 47Settings
- 62Pricing
- 79Article
- Rank 4, could be anywhere from 1 to 11Claude Sonnet 5.5NewAnthropic · closed weights · $0.14 per page
- 62Landing
- 57Dashboard
- 56Settings
- 56Pricing
- 78Article
- Rank 5, could be anywhere from 2 to 10GPT-6 LunaNewOpenAI · closed weights · $0.005 per page
- 70Landing
- 61Dashboard
- 53Settings
- 52Pricing
- 74Article
- Rank 6, could be anywhere from 1 to 11Gemini 4 ArgonNewGoogle · closed weights · $0.15 per page
- 66Landing
- 55Dashboard
- 54Settings
- 56Pricing
- 78Article
- Rank 7, could be anywhere from 1 to 12Muse WonderNewMeta · closed weights · $0.25 per page
- 63Landing
- 62Dashboard
- 55Settings
- 49Pricing
- 74Article
- Rank 8, could be anywhere from 2 to 13Claude Opus 5.5Anthropic · closed weights · $0.28 per page
- 63Landing
- 59Dashboard
- 52Settings
- 51Pricing
- 78Article
- Rank 9, could be anywhere from 3 to 11GPT-5.6 SolOpenAI · closed weights · $0.17 per page
- 67Landing
- 52Dashboard
- 53Settings
- 51Pricing
- 72Article
- Rank 10, could be anywhere from 5 to 13Claude Fable 5.1Anthropic · closed weights · $0.87 per page
- 68Landing
- 58Dashboard
- 51Settings
- 50Pricing
- 68Article
- Rank 11, could be anywhere from 5 to 13Claude Fable 5Anthropic · closed weights · $0.53 per page
- 60Landing
- 56Dashboard
- 49Settings
- 50Pricing
- 74Article
- Rank 12, could be anywhere from 8 to 14Gemini 3.8 FlashNewGoogle · closed weights · $0.056 per page
- 64Landing
- 50Dashboard
- 50Settings
- 50Pricing
- 63Article
- Rank 13, could be anywhere from 8 to 14DeepSeek V4 ProNewDeepSeek · open weights for DeepSeek V4 Pro · $0.032 per page
- 60Landing
- 49Dashboard
- 48Settings
- 51Pricing
- 65Article
- Rank 14, could be anywhere from 12 to 14Claude Opus 4.8Anthropic · closed weights · $0.15 per page
- 52Landing
- 49Dashboard
- 47Settings
- 49Pricing
- 68Article
Unranked · no scored page on every brief
- UnrankedQwen 3.8 MaxNewAlibaba · closed weights · $0.16 per page3 of 10 pages were cut off before they finished. No Dashboard build finished, and a model needs a scored page on all five briefs to be ranked.
- UnrankedKimi K3NewMoonshot · open weights for Kimi K3 · $0.41 per page3 of 10 pages were cut off before they finished. No Dashboard build finished, and a model needs a scored page on all five briefs to be ranked.
- UnrankedGLM 5.3NewZhipu · open weights for GLM 5.3 · $0.14 per page9 of 10 pages broke: 8 came back with no page and 1 was cut off before it finished. No Landing, Dashboard, Settings or Pricing build finished, and a model needs a scored page on all five briefs to be ranked.
Too close to call at the top. GPT-6 Astra has the highest Index, 66.4, but it is not clearly ahead of Grok 4.7, GPT-6.1 Sol, Claude Sonnet 5.5, Gemini 4 Argon or Muse Wonder: the 95% interval on each gap includes zero, so their order is not proven. 37 of the 91 head-to-head pairs are clearly apart. The other 54 sit inside the noise.
- Models
- 14 + 3 unranked
- Pages built
- 155
- Judge passes
- 170
- Briefs
- 5 + 0 private
- Viewports
- 1440 · 390 px
- Engine
- f0bb659
Leaderboard
Rows are numbered by Rams Index, 1 for the highest. Under each number is its range: every place the paired 95% intervals allow, from 1 plus the models clearly better to the total minus the models clearly worse.
1GPT-6 AstraNewOpenAI · closed weights · 2026-0966.457.7–75.0
Provisional: 1.0 judge passes per page, protocol 2
GPT-6 Astra: grades, best and worst page
Cost and checks
- $ per page
- $0.43
- Grade sum, 7–35
- 25.60
- Terra, cross-judge
- 80.7
- Visual, secondary
- 90.7
- Defects per page
- 6.5
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 113s
- Output tokens (reasoning), median
- 7.7k (0.5k)
- Broken pages
- 0 of 10
- Pages with a critical
- 90%
Index by brief
- Landing
- 72.3
- Dashboard
- 62.5
- Settings
- 50.9
- Pricing
- 65.2
- Article
- 81.2
Grades by dimension, 1–5
- Hierarchy
- 3.85
- Type
- 3.50
- Spacing
- 3.30
- Color
- 3.65
- Components
- 3.80
- Composure
- 3.80
- Craft
- 3.70
2Grok 4.7NewSpaceXSI · closed weights · 2026-0966.158.6–72.0
Provisional: 1.0 judge passes per page, protocol 2
Grok 4.7: grades, best and worst page
Cost and checks
- $ per page
- $0.13
- Grade sum, 7–35
- 25.50
- Terra, cross-judge
- 84.5
- Visual, secondary
- 88.6
- Defects per page
- 8.0
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 284s
- Output tokens (reasoning), median
- 22.1k (13.1k)
- Broken pages
- 0 of 10
- Pages with a critical
- 0%
Index by brief
- Landing
- 69.6
- Dashboard
- 65.2
- Settings
- 51.8
- Pricing
- 68.8
- Article
- 75.0
Grades by dimension, 1–5
- Hierarchy
- 3.70
- Type
- 3.75
- Spacing
- 3.30
- Color
- 3.80
- Components
- 3.55
- Composure
- 3.65
- Craft
- 3.75
3GPT-6.1 SolOpenAI · closed weights · 2026-0965.455.7–73.9
Provisional: 1.0 judge passes per page, protocol 2
GPT-6.1 Sol: grades, best and worst page
Cost and checks
- $ per page
- $0.15
- Grade sum, 7–35
- 25.30
- Terra, cross-judge
- 79.8
- Visual, secondary
- 88.8
- Defects per page
- 5.2
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 102s
- Output tokens (reasoning), median
- 7.2k (0.4k)
- Broken pages
- 0 of 10
- Pages with a critical
- 80%
Index by brief
- Landing
- 70.5
- Dashboard
- 68.8
- Settings
- 47.3
- Pricing
- 61.6
- Article
- 78.6
Grades by dimension, 1–5
- Hierarchy
- 3.80
- Type
- 3.65
- Spacing
- 3.25
- Color
- 3.70
- Components
- 3.60
- Composure
- 3.60
- Craft
- 3.70
4Claude Sonnet 5.5NewAnthropic · closed weights · 2026-0961.856.1–70.0
Provisional: 1.0 judge passes per page, protocol 2
Claude Sonnet 5.5: grades, best and worst page
Cost and checks
- $ per page
- $0.14
- Grade sum, 7–35
- 24.30
- Terra, cross-judge
- 74.5
- Visual, secondary
- 90.2
- Defects per page
- 2.6
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 95s
- Output tokens (reasoning), median
- 13.6k (2.7k)
- Broken pages
- 0 of 10
- Pages with a critical
- 10%
Index by brief
- Landing
- 61.6
- Dashboard
- 57.1
- Settings
- 56.2
- Pricing
- 56.2
- Article
- 77.7
Grades by dimension, 1–5
- Hierarchy
- 3.75
- Type
- 3.40
- Spacing
- 3.20
- Color
- 3.35
- Components
- 3.55
- Composure
- 3.75
- Craft
- 3.30
5GPT-6 LunaNewOpenAI · closed weights · 2026-0961.854.5–69.5
Provisional: 1.0 judge passes per page, protocol 2
GPT-6 Luna: grades, best and worst page
Cost and checks
- $ per page
- $0.005
- Grade sum, 7–35
- 24.30
- Terra, cross-judge
- 78.2
- Visual, secondary
- 88.3
- Defects per page
- 5.3
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 81s
- Output tokens (reasoning), median
- 9.2k (0.8k)
- Broken pages
- 0 of 10
- Pages with a critical
- 70%
Index by brief
- Landing
- 69.6
- Dashboard
- 60.7
- Settings
- 52.7
- Pricing
- 51.8
- Article
- 74.1
Grades by dimension, 1–5
- Hierarchy
- 3.65
- Type
- 3.50
- Spacing
- 3.15
- Color
- 3.55
- Components
- 3.65
- Composure
- 3.30
- Craft
- 3.50
6Gemini 4 ArgonNewGoogle · closed weights · 2026-1061.654.5–70.9
Provisional: 1.0 judge passes per page, protocol 2
Gemini 4 Argon: grades, best and worst page
Cost and checks
- $ per page
- $0.15
- Grade sum, 7–35
- 24.25
- Terra, cross-judge
- 76.1
- Visual, secondary
- 86.4
- Defects per page
- 3.1
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 171s
- Output tokens (reasoning), median
- 15.7k (0.0k)
- Broken pages
- 0 of 10
- Pages with a critical
- 50%
Index by brief
- Landing
- 66.1
- Dashboard
- 54.5
- Settings
- 53.6
- Pricing
- 56.2
- Article
- 77.7
Grades by dimension, 1–5
- Hierarchy
- 3.70
- Type
- 3.35
- Spacing
- 3.15
- Color
- 3.55
- Components
- 3.40
- Composure
- 3.65
- Craft
- 3.45
7Muse WonderNewMeta · closed weights · 2026-1060.553.2–68.9
Provisional: 1.0 judge passes per page, protocol 2
Muse Wonder: grades, best and worst page
Cost and checks
- $ per page
- $0.25
- Grade sum, 7–35
- 23.95
- Terra, cross-judge
- 74.8
- Visual, secondary
- 86.0
- Defects per page
- 8.5
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 125s
- Output tokens (reasoning), median
- 12.4k (1.1k)
- Broken pages
- 0 of 10
- Pages with a critical
- 20%
Index by brief
- Landing
- 63.4
- Dashboard
- 61.6
- Settings
- 54.5
- Pricing
- 49.1
- Article
- 74.1
Grades by dimension, 1–5
- Hierarchy
- 3.50
- Type
- 3.35
- Spacing
- 3.15
- Color
- 3.55
- Components
- 3.50
- Composure
- 3.50
- Craft
- 3.40
8Claude Opus 5.5Anthropic · closed weights · 2026-0960.452.9–70.2
Provisional: 1.0 judge passes per page, protocol 2
Claude Opus 5.5: grades, best and worst page
Cost and checks
- $ per page
- $0.28
- Grade sum, 7–35
- 23.90
- Terra, cross-judge
- 70.7
- Visual, secondary
- 86.5
- Defects per page
- 3.2
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default (medium)
- Time per page, median
- 102s
- Output tokens (reasoning), median
- 13.2k (n/a)
- Broken pages
- 0 of 10
- Pages with a critical
- 10%
Index by brief
- Landing
- 62.5
- Dashboard
- 58.9
- Settings
- 51.8
- Pricing
- 50.9
- Article
- 77.7
Grades by dimension, 1–5
- Hierarchy
- 3.60
- Type
- 3.40
- Spacing
- 3.15
- Color
- 3.50
- Components
- 3.40
- Composure
- 3.45
- Craft
- 3.40
9GPT-5.6 SolOpenAI · closed weights · 2026-0658.951.6–66.9
Provisional: 1.3 judge passes per page, protocol 2
GPT-5.6 Sol: grades, best and worst page
Cost and checks
- $ per page
- $0.17
- Grade sum, 7–35
- 23.50
- Terra, cross-judge
- 79.3
- Visual, secondary
- 88.6
- Defects per page
- 4.7
- Pages · passes
- 15 · 20
How it ran
- Effort
- Default
- Time per page, median
- 57s
- Output tokens (reasoning), median
- 8.5k (0.2k)
- Broken pages
- 0 of 15
- Pages with a critical
- 33%
Index by brief
- Landing
- 67.0
- Dashboard
- 52.1
- Settings
- 52.7
- Pricing
- 51.2
- Article
- 71.7
Grades by dimension, 1–5
- Hierarchy
- 3.40
- Type
- 3.20
- Spacing
- 3.18
- Color
- 3.33
- Components
- 3.57
- Composure
- 3.43
- Craft
- 3.38
10Claude Fable 5.1Anthropic · closed weights · 2026-0858.952.0–65.9
Provisional: 1.0 judge passes per page, protocol 2
Claude Fable 5.1: grades, best and worst page
Cost and checks
- $ per page
- $0.87
- Grade sum, 7–35
- 23.50
- Terra, cross-judge
- 77.1
- Visual, secondary
- 88.5
- Defects per page
- 3.8
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default (always thinks)
- Time per page, median
- 180s
- Output tokens (reasoning), median
- 17.3k (n/a)
- Broken pages
- 0 of 10
- Pages with a critical
- 30%
Index by brief
- Landing
- 67.9
- Dashboard
- 58.0
- Settings
- 50.9
- Pricing
- 50.0
- Article
- 67.9
Grades by dimension, 1–5
- Hierarchy
- 3.45
- Type
- 3.35
- Spacing
- 3.10
- Color
- 3.35
- Components
- 3.35
- Composure
- 3.50
- Craft
- 3.40
11Claude Fable 5Anthropic · closed weights · 2026-0657.651.0–67.0
Provisional: 1.3 judge passes per page, protocol 2
Claude Fable 5: grades, best and worst page
Cost and checks
- $ per page
- $0.53
- Grade sum, 7–35
- 23.13
- Terra, cross-judge
- 74.5
- Visual, secondary
- 88.7
- Defects per page
- 3.2
- Pages · passes
- 15 · 20
How it ran
- Effort
- Default (always thinks)
- Time per page, median
- 99s
- Output tokens (reasoning), median
- 10.3k (n/a)
- Broken pages
- 0 of 15
- Pages with a critical
- 20%
Index by brief
- Landing
- 59.5
- Dashboard
- 56.0
- Settings
- 49.4
- Pricing
- 49.7
- Article
- 73.5
Grades by dimension, 1–5
- Hierarchy
- 3.33
- Type
- 3.28
- Spacing
- 3.20
- Color
- 3.27
- Components
- 3.35
- Composure
- 3.38
- Craft
- 3.32
12Gemini 3.8 FlashNewGoogle · closed weights · 2026-0955.550.0–62.0
Provisional: 1.0 judge passes per page, protocol 2
Gemini 3.8 Flash: grades, best and worst page
Cost and checks
- $ per page
- $0.056
- Grade sum, 7–35
- 22.55
- Terra, cross-judge
- 69.6
- Visual, secondary
- 84.2
- Defects per page
- 5.0
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 94s
- Output tokens (reasoning), median
- 14.4k (1.6k)
- Broken pages
- 0 of 10
- Pages with a critical
- 40%
Index by brief
- Landing
- 64.3
- Dashboard
- 50.0
- Settings
- 50.0
- Pricing
- 50.0
- Article
- 63.4
Grades by dimension, 1–5
- Hierarchy
- 3.15
- Type
- 3.25
- Spacing
- 3.10
- Color
- 3.35
- Components
- 3.35
- Composure
- 3.05
- Craft
- 3.30
13DeepSeek V4 ProNewDeepSeek · open weights · 2026-0854.648.9–61.4
Provisional: 1.0 judge passes per page, protocol 2
DeepSeek V4 Pro: grades, best and worst page
Cost and checks
- $ per page
- $0.032
- Grade sum, 7–35
- 22.30
- Terra, cross-judge
- 69.8
- Visual, secondary
- 87.3
- Defects per page
- 3.2
- Pages · passes
- 10 · 10
How it ran
- Effort
- Default
- Time per page, median
- 196s
- Output tokens (reasoning), median
- 15.9k (5.8k)
- Broken pages
- 0 of 10
- Pages with a critical
- 10%
Index by brief
- Landing
- 59.8
- Dashboard
- 49.1
- Settings
- 48.2
- Pricing
- 50.9
- Article
- 65.2
Grades by dimension, 1–5
- Hierarchy
- 3.25
- Type
- 3.10
- Spacing
- 3.05
- Color
- 3.20
- Components
- 3.35
- Composure
- 3.15
- Craft
- 3.20
14Claude Opus 4.8Anthropic · closed weights · 2026-0552.948.0–60.5
Provisional: 1.3 judge passes per page, protocol 2
Claude Opus 4.8: grades, best and worst page
Cost and checks
- $ per page
- $0.15
- Grade sum, 7–35
- 21.80
- Terra, cross-judge
- 69.5
- Visual, secondary
- 85.2
- Defects per page
- 2.5
- Pages · passes
- 15 · 20
How it ran
- Effort
- Default
- Time per page, median
- 61s
- Output tokens (reasoning), median
- 6.2k (n/a)
- Broken pages
- 0 of 15
- Pages with a critical
- 33%
Index by brief
- Landing
- 51.5
- Dashboard
- 48.8
- Settings
- 47.3
- Pricing
- 49.1
- Article
- 67.6
Grades by dimension, 1–5
- Hierarchy
- 3.25
- Type
- 3.10
- Spacing
- 3.05
- Color
- 3.05
- Components
- 3.02
- Composure
- 3.17
- Craft
- 3.17
Unranked · no scored page on every brief
UnrankedQwen 3.8 MaxNewAlibaba · closed weights3 of 10 pages were cut off before they finished.
3 of 10 pages were cut off before they finished. No Dashboard build finished, and a model needs a scored page on all five briefs to be ranked.
Qwen 3.8 Max: why it has no rank
How it ran
- Effort
- Default
- Time per page, median
- 645s
- Output tokens (reasoning), median
- 25.0k (11.0k)
- Broken pages
- 3 of 10
- Pages with a critical
- 43% (3 of 7)
Index by brief
- Landing
- 76.8
- Dashboard
- Broken
- Settings
- 59.8
- Pricing
- 67.9
- Article
- 75.0
Cost
- $ per page
- $0.16
UnrankedKimi K3NewMoonshot · open weights3 of 10 pages were cut off before they finished.
3 of 10 pages were cut off before they finished. No Dashboard build finished, and a model needs a scored page on all five briefs to be ranked.
How it ran
- Effort
- Default
- Time per page, median
- 721s
- Output tokens (reasoning), median
- 28.4k (17.2k)
- Broken pages
- 3 of 10
- Pages with a critical
- 0% (0 of 7)
Index by brief
- Landing
- 68.8
- Dashboard
- Broken
- Settings
- 55.4
- Pricing
- 58.9
- Article
- 75.9
Cost
- $ per page
- $0.41
UnrankedGLM 5.3NewZhipu · open weights9 of 10 pages broke: 8 came back with no page and 1 was cut off before it finished.
9 of 10 pages broke: 8 came back with no page and 1 was cut off before it finished. No Landing, Dashboard, Settings or Pricing build finished, and a model needs a scored page on all five briefs to be ranked.
How it ran
- Effort
- Default
- Time per page, median
- 542s
- Output tokens (reasoning), median
- 32.0k (31.9k)
- Broken pages
- 9 of 10
- Pages with a critical
- 0% (0 of 1)
Index by brief
- Landing
- Broken
- Dashboard
- Broken
- Settings
- Broken
- Pricing
- Broken
- Article
- 73.2
Cost
- $ per page
- $0.14
| 0–100 | 45–75 | Generation | Provider default | Median per page | Median (reasoning) | Pages | Share of pages | 7–35 | Index | Index | Index | Index | Index | Cross-judge | Secondary | Per page | Passes | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1, could be anywhere from 1 to 6 | GPT-6 AstraProvisional: 1.0 judge passes per page, protocol 2NewOpenAI · closed weights · 2026-09 | 66.4 | 57.7–75.0 | $0.43 | Default | 113s | 7.7k (0.5k) | 0 of 10 | 90% | 25.60 | 72.3 | 62.5 | 50.9 | 65.2 | 81.2 | 80.7 | 90.7 | 6.5 | 10 · 10 |
| 2, could be anywhere from 1 to 8 | Grok 4.7Provisional: 1.0 judge passes per page, protocol 2NewSpaceXSI · closed weights · 2026-09 | 66.1 | 58.6–72.0 | $0.13 | Default | 284s | 22.1k (13.1k) | 0 of 10 | 0% | 25.50 | 69.6 | 65.2 | 51.8 | 68.8 | 75.0 | 84.5 | 88.6 | 8.0 | 10 · 10 |
| 3, could be anywhere from 1 to 9 | GPT-6.1 SolProvisional: 1.0 judge passes per page, protocol 2OpenAI · closed weights · 2026-09 | 65.4 | 55.7–73.9 | $0.15 | Default | 102s | 7.2k (0.4k) | 0 of 10 | 80% | 25.30 | 70.5 | 68.8 | 47.3 | 61.6 | 78.6 | 79.8 | 88.8 | 5.2 | 10 · 10 |
| 4, could be anywhere from 1 to 11 | Claude Sonnet 5.5Provisional: 1.0 judge passes per page, protocol 2NewAnthropic · closed weights · 2026-09 | 61.8 | 56.1–70.0 | $0.14 | Default | 95s | 13.6k (2.7k) | 0 of 10 | 10% | 24.30 | 61.6 | 57.1 | 56.2 | 56.2 | 77.7 | 74.5 | 90.2 | 2.6 | 10 · 10 |
| 5, could be anywhere from 2 to 10 | GPT-6 LunaProvisional: 1.0 judge passes per page, protocol 2NewOpenAI · closed weights · 2026-09 | 61.8 | 54.5–69.5 | $0.005 | Default | 81s | 9.2k (0.8k) | 0 of 10 | 70% | 24.30 | 69.6 | 60.7 | 52.7 | 51.8 | 74.1 | 78.2 | 88.3 | 5.3 | 10 · 10 |
| 6, could be anywhere from 1 to 11 | Gemini 4 ArgonProvisional: 1.0 judge passes per page, protocol 2NewGoogle · closed weights · 2026-10 | 61.6 | 54.5–70.9 | $0.15 | Default | 171s | 15.7k (0.0k) | 0 of 10 | 50% | 24.25 | 66.1 | 54.5 | 53.6 | 56.2 | 77.7 | 76.1 | 86.4 | 3.1 | 10 · 10 |
| 7, could be anywhere from 1 to 12 | Muse WonderProvisional: 1.0 judge passes per page, protocol 2NewMeta · closed weights · 2026-10 | 60.5 | 53.2–68.9 | $0.25 | Default | 125s | 12.4k (1.1k) | 0 of 10 | 20% | 23.95 | 63.4 | 61.6 | 54.5 | 49.1 | 74.1 | 74.8 | 86.0 | 8.5 | 10 · 10 |
| 8, could be anywhere from 2 to 13 | Claude Opus 5.5Provisional: 1.0 judge passes per page, protocol 2Anthropic · closed weights · 2026-09 | 60.4 | 52.9–70.2 | $0.28 | Default (medium) | 102s | 13.2k (n/a) | 0 of 10 | 10% | 23.90 | 62.5 | 58.9 | 51.8 | 50.9 | 77.7 | 70.7 | 86.5 | 3.2 | 10 · 10 |
| 9, could be anywhere from 3 to 11 | GPT-5.6 SolProvisional: 1.3 judge passes per page, protocol 2OpenAI · closed weights · 2026-06 | 58.9 | 51.6–66.9 | $0.17 | Default | 57s | 8.5k (0.2k) | 0 of 15 | 33% | 23.50 | 67.0 | 52.1 | 52.7 | 51.2 | 71.7 | 79.3 | 88.6 | 4.7 | 15 · 20 |
| 10, could be anywhere from 5 to 13 | Claude Fable 5.1Provisional: 1.0 judge passes per page, protocol 2Anthropic · closed weights · 2026-08 | 58.9 | 52.0–65.9 | $0.87 | Default (always thinks) | 180s | 17.3k (n/a) | 0 of 10 | 30% | 23.50 | 67.9 | 58.0 | 50.9 | 50.0 | 67.9 | 77.1 | 88.5 | 3.8 | 10 · 10 |
| 11, could be anywhere from 5 to 13 | Claude Fable 5Provisional: 1.3 judge passes per page, protocol 2Anthropic · closed weights · 2026-06 | 57.6 | 51.0–67.0 | $0.53 | Default (always thinks) | 99s | 10.3k (n/a) | 0 of 15 | 20% | 23.13 | 59.5 | 56.0 | 49.4 | 49.7 | 73.5 | 74.5 | 88.7 | 3.2 | 15 · 20 |
| 12, could be anywhere from 8 to 14 | Gemini 3.8 FlashProvisional: 1.0 judge passes per page, protocol 2NewGoogle · closed weights · 2026-09 | 55.5 | 50.0–62.0 | $0.056 | Default | 94s | 14.4k (1.6k) | 0 of 10 | 40% | 22.55 | 64.3 | 50.0 | 50.0 | 50.0 | 63.4 | 69.6 | 84.2 | 5.0 | 10 · 10 |
| 13, could be anywhere from 8 to 14 | DeepSeek V4 ProProvisional: 1.0 judge passes per page, protocol 2NewDeepSeek · open weights for DeepSeek V4 Pro · 2026-08 | 54.6 | 48.9–61.4 | $0.032 | Default | 196s | 15.9k (5.8k) | 0 of 10 | 10% | 22.30 | 59.8 | 49.1 | 48.2 | 50.9 | 65.2 | 69.8 | 87.3 | 3.2 | 10 · 10 |
| 14, could be anywhere from 12 to 14 | Claude Opus 4.8Provisional: 1.3 judge passes per page, protocol 2Anthropic · closed weights · 2026-05 | 52.9 | 48.0–60.5 | $0.15 | Default | 61s | 6.2k (n/a) | 0 of 15 | 33% | 21.80 | 51.5 | 48.8 | 47.3 | 49.1 | 67.6 | 69.5 | 85.2 | 2.5 | 15 · 20 |
| Unranked · no scored page on every brief, so no Index or interval | |||||||||||||||||||
| Unranked | Qwen 3.8 MaxNewAlibaba · closed weights · 2026-093 of 10 pages were cut off before they finished. | – | – | $0.16 | Default | 645s | 25.0k (11.0k) | 3 of 10 | 43% (3 of 7) | – | 76.8 | Broken | 59.8 | 67.9 | 75.0 | – | – | – | 7 · – |
| Unranked | Kimi K3NewMoonshot · open weights for Kimi K3 · 2026-073 of 10 pages were cut off before they finished. | – | – | $0.41 | Default | 721s | 28.4k (17.2k) | 3 of 10 | 0% (0 of 7) | – | 68.8 | Broken | 55.4 | 58.9 | 75.9 | – | – | – | 7 · – |
| Unranked | GLM 5.3NewZhipu · open weights for GLM 5.3 · 2026-089 of 10 pages broke: 8 came back with no page and 1 was cut off before it finished. | – | – | $0.14 | Default | 542s | 32.0k (31.9k) | 9 of 10 | 0% (0 of 1) | – | Broken | Broken | Broken | Broken | 73.2 | – | – | – | 1 · – |
- Rams Index = (grade sum − 7) / 28 × 100. The grade sum is seven 1–5 grades from Rams eyes · Claude Sonnet 4.6, averaged over desktop and phone, then passes, then briefs.
- * Provisional, every row in v0.5: fewer judge passes per page, protocol 2 (v0.5 ran 1.0–1.3). Tap a star for the model's reason. A starred rank is never quoted.
- Terra is the same pipeline with GPT-5.6 Terra as the judge, so a reader can check whether a Claude judge flatters Claude pages. Visual is the judge's finding-based score, secondary.
- Harness: one single-file HTML page per build, the same brief text for every model, every build judged as rendered.
- Effort: every model runs at its provider’s default. We send no effort or thinking setting; where the default is known it is named. Time and output tokens are medians per page, over every build, broken ones included. Reasoning tokens (in parentheses) are what each API reports; Anthropic’s doesn’t split them out, so those rows say n/a.
- Broken: builds that came back with no page, or a page cut off before it finished. A broken page is counted and never scored. Criticals: the share of scored pages whose first Rams code review found a critical issue.
Frontier
Rams Index against generation cost per page, log scale.
A model is on the frontier when no cheaper model is clearly better. Clearly better uses the same paired interval as the rank, so a cheaper model has to beat it outright, not by a point inside the noise.
Cost is what the model's API charged to build the page, from the run ledger at list price on run day. Judging cost is ours and stays out.
- GPT-6 Luna$0.005 · 61.8 · frontier
- DeepSeek V4 Pro$0.032 · 54.6 · beaten by a cheaper model
- Gemini 3.8 Flash$0.056 · 55.5 · beaten by a cheaper model
- Grok 4.7$0.13 · 66.1 · frontier
- Claude Sonnet 5.5$0.14 · 61.8 · frontier
- Gemini 4 Argon$0.15 · 61.6 · frontier
- GPT-6.1 Sol$0.15 · 65.4 · frontier
- Claude Opus 4.8$0.15 · 52.9 · beaten by a cheaper model
- GPT-5.6 Sol$0.17 · 58.9 · beaten by a cheaper model
- Muse Wonder$0.25 · 60.5 · frontier
- Claude Opus 5.5$0.28 · 60.4 · frontier
- GPT-6 Astra$0.43 · 66.4 · frontier
- Claude Fable 5$0.53 · 57.6 · beaten by a cheaper model
- Claude Fable 5.1$0.87 · 58.9 · beaten by a cheaper model
Head to head
Row model against column model.Each model against every model below it. The number is the paired difference in Index points.
| GPT-6 Astra | Grok 4.7 | GPT-6.1 Sol | Sonnet 5.5 | GPT-6 Luna | Gemini 4 Argon | Muse Wonder | Opus 5.5 | GPT-5.6 Sol | Fable 5.1 | Fable 5 | Gemini 3.8 Flash | DeepSeek V4 Pro | Opus 4.8 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT-6 Astra | +0.4, inside the noise | +1.1, inside the noise | +4.6, inside the noise | +4.6, clearly better | +4.8, inside the noise | +5.9, inside the noise | +6.1, clearly better | +7.5, clearly better | +7.5, clearly better | +8.8, clearly better | +10.9, clearly better | +11.8, clearly better | +13.6, clearly better | |
| Grok 4.7 | -0.4, inside the noise | +0.7, inside the noise | +4.3, inside the noise | +4.3, inside the noise | +4.5, inside the noise | +5.5, inside the noise | +5.7, inside the noise | +7.1, clearly better | +7.1, clearly better | +8.5, clearly better | +10.5, clearly better | +11.4, clearly better | +13.2, clearly better | |
| GPT-6.1 Sol | -1.1, inside the noise | -0.7, inside the noise | +3.6, inside the noise | +3.6, inside the noise | +3.8, inside the noise | +4.8, inside the noise | +5.0, inside the noise | +6.4, inside the noise | +6.4, clearly better | +7.8, clearly better | +9.8, clearly better | +10.7, clearly better | +12.5, clearly better | |
| Claude Sonnet 5.5 | -4.6, inside the noise | -4.3, inside the noise | -3.6, inside the noise | 0.0, inside the noise | +0.2, inside the noise | +1.2, inside the noise | +1.4, inside the noise | +2.9, inside the noise | +2.9, inside the noise | +4.2, clearly better | +6.2, inside the noise | +7.1, clearly better | +8.9, clearly better | |
| GPT-6 Luna | -4.6, clearly behind | -4.3, inside the noise | -3.6, inside the noise | 0.0, inside the noise | +0.2, inside the noise | +1.2, inside the noise | +1.4, inside the noise | +2.9, inside the noise | +2.9, clearly better | +4.2, inside the noise | +6.2, clearly better | +7.1, clearly better | +8.9, clearly better | |
| Gemini 4 Argon | -4.8, inside the noise | -4.5, inside the noise | -3.8, inside the noise | -0.2, inside the noise | -0.2, inside the noise | +1.1, inside the noise | +1.2, inside the noise | +2.7, inside the noise | +2.7, inside the noise | +4.0, inside the noise | +6.1, clearly better | +7.0, clearly better | +8.8, clearly better | |
| Muse Wonder | -5.9, inside the noise | -5.5, inside the noise | -4.8, inside the noise | -1.2, inside the noise | -1.2, inside the noise | -1.1, inside the noise | +0.2, inside the noise | +1.6, inside the noise | +1.6, inside the noise | +2.9, inside the noise | +5.0, clearly better | +5.9, inside the noise | +7.7, clearly better | |
| Claude Opus 5.5 | -6.1, clearly behind | -5.7, inside the noise | -5.0, inside the noise | -1.4, inside the noise | -1.4, inside the noise | -1.2, inside the noise | -0.2, inside the noise | +1.4, inside the noise | +1.4, inside the noise | +2.8, inside the noise | +4.8, inside the noise | +5.7, inside the noise | +7.5, clearly better | |
| GPT-5.6 Sol | -7.5, clearly behind | -7.1, clearly behind | -6.4, inside the noise | -2.9, inside the noise | -2.9, inside the noise | -2.7, inside the noise | -1.6, inside the noise | -1.4, inside the noise | 0.0, inside the noise | +1.3, inside the noise | +3.4, clearly better | +4.3, clearly better | +6.1, clearly better | |
| Claude Fable 5.1 | -7.5, clearly behind | -7.1, clearly behind | -6.4, clearly behind | -2.9, inside the noise | -2.9, clearly behind | -2.7, inside the noise | -1.6, inside the noise | -1.4, inside the noise | 0.0, inside the noise | +1.3, inside the noise | +3.4, inside the noise | +4.3, inside the noise | +6.1, clearly better | |
| Claude Fable 5 | -8.8, clearly behind | -8.5, clearly behind | -7.8, clearly behind | -4.2, clearly behind | -4.2, inside the noise | -4.0, inside the noise | -2.9, inside the noise | -2.8, inside the noise | -1.3, inside the noise | -1.3, inside the noise | +2.1, inside the noise | +3.0, inside the noise | +4.8, clearly better | |
| Gemini 3.8 Flash | -10.9, clearly behind | -10.5, clearly behind | -9.8, clearly behind | -6.2, inside the noise | -6.2, clearly behind | -6.1, clearly behind | -5.0, clearly behind | -4.8, inside the noise | -3.4, clearly behind | -3.4, inside the noise | -2.1, inside the noise | +0.9, inside the noise | +2.7, inside the noise | |
| DeepSeek V4 Pro | -11.8, clearly behind | -11.4, clearly behind | -10.7, clearly behind | -7.1, clearly behind | -7.1, clearly behind | -7.0, clearly behind | -5.9, inside the noise | -5.7, inside the noise | -4.3, clearly behind | -4.3, inside the noise | -3.0, inside the noise | -0.9, inside the noise | +1.8, inside the noise | |
| Claude Opus 4.8 | -13.6, clearly behind | -13.2, clearly behind | -12.5, clearly behind | -8.9, clearly behind | -8.9, clearly behind | -8.8, clearly behind | -7.7, clearly behind | -7.5, clearly behind | -6.1, clearly behind | -6.1, clearly behind | -4.8, clearly behind | -2.7, inside the noise | -1.8, inside the noise |
GPT-6 Astra
- vs Grok 4.7+0.4, inside the noise
- vs GPT-6.1 Sol+1.1, inside the noise
- vs Claude Sonnet 5.5+4.6, inside the noise
- vs GPT-6 Luna+4.6, clearly better
- vs Gemini 4 Argon+4.8, inside the noise
- vs Muse Wonder+5.9, inside the noise
- vs Claude Opus 5.5+6.1, clearly better
- vs GPT-5.6 Sol+7.5, clearly better
- vs Claude Fable 5.1+7.5, clearly better
- vs Claude Fable 5+8.8, clearly better
- vs Gemini 3.8 Flash+10.9, clearly better
- vs DeepSeek V4 Pro+11.8, clearly better
- vs Claude Opus 4.8+13.6, clearly better
Grok 4.7
- vs GPT-6.1 Sol+0.7, inside the noise
- vs Claude Sonnet 5.5+4.3, inside the noise
- vs GPT-6 Luna+4.3, inside the noise
- vs Gemini 4 Argon+4.5, inside the noise
- vs Muse Wonder+5.5, inside the noise
- vs Claude Opus 5.5+5.7, inside the noise
- vs GPT-5.6 Sol+7.1, clearly better
- vs Claude Fable 5.1+7.1, clearly better
- vs Claude Fable 5+8.5, clearly better
- vs Gemini 3.8 Flash+10.5, clearly better
- vs DeepSeek V4 Pro+11.4, clearly better
- vs Claude Opus 4.8+13.2, clearly better
GPT-6.1 Sol
- vs Claude Sonnet 5.5+3.6, inside the noise
- vs GPT-6 Luna+3.6, inside the noise
- vs Gemini 4 Argon+3.8, inside the noise
- vs Muse Wonder+4.8, inside the noise
- vs Claude Opus 5.5+5.0, inside the noise
- vs GPT-5.6 Sol+6.4, inside the noise
- vs Claude Fable 5.1+6.4, clearly better
- vs Claude Fable 5+7.8, clearly better
- vs Gemini 3.8 Flash+9.8, clearly better
- vs DeepSeek V4 Pro+10.7, clearly better
- vs Claude Opus 4.8+12.5, clearly better
Claude Sonnet 5.5
- vs GPT-6 Luna0.0, inside the noise
- vs Gemini 4 Argon+0.2, inside the noise
- vs Muse Wonder+1.2, inside the noise
- vs Claude Opus 5.5+1.4, inside the noise
- vs GPT-5.6 Sol+2.9, inside the noise
- vs Claude Fable 5.1+2.9, inside the noise
- vs Claude Fable 5+4.2, clearly better
- vs Gemini 3.8 Flash+6.2, inside the noise
- vs DeepSeek V4 Pro+7.1, clearly better
- vs Claude Opus 4.8+8.9, clearly better
GPT-6 Luna
- vs Gemini 4 Argon+0.2, inside the noise
- vs Muse Wonder+1.2, inside the noise
- vs Claude Opus 5.5+1.4, inside the noise
- vs GPT-5.6 Sol+2.9, inside the noise
- vs Claude Fable 5.1+2.9, clearly better
- vs Claude Fable 5+4.2, inside the noise
- vs Gemini 3.8 Flash+6.2, clearly better
- vs DeepSeek V4 Pro+7.1, clearly better
- vs Claude Opus 4.8+8.9, clearly better
Gemini 4 Argon
- vs Muse Wonder+1.1, inside the noise
- vs Claude Opus 5.5+1.2, inside the noise
- vs GPT-5.6 Sol+2.7, inside the noise
- vs Claude Fable 5.1+2.7, inside the noise
- vs Claude Fable 5+4.0, inside the noise
- vs Gemini 3.8 Flash+6.1, clearly better
- vs DeepSeek V4 Pro+7.0, clearly better
- vs Claude Opus 4.8+8.8, clearly better
Muse Wonder
- vs Claude Opus 5.5+0.2, inside the noise
- vs GPT-5.6 Sol+1.6, inside the noise
- vs Claude Fable 5.1+1.6, inside the noise
- vs Claude Fable 5+2.9, inside the noise
- vs Gemini 3.8 Flash+5.0, clearly better
- vs DeepSeek V4 Pro+5.9, inside the noise
- vs Claude Opus 4.8+7.7, clearly better
Claude Opus 5.5
- vs GPT-5.6 Sol+1.4, inside the noise
- vs Claude Fable 5.1+1.4, inside the noise
- vs Claude Fable 5+2.8, inside the noise
- vs Gemini 3.8 Flash+4.8, inside the noise
- vs DeepSeek V4 Pro+5.7, inside the noise
- vs Claude Opus 4.8+7.5, clearly better
GPT-5.6 Sol
- vs Claude Fable 5.10.0, inside the noise
- vs Claude Fable 5+1.3, inside the noise
- vs Gemini 3.8 Flash+3.4, clearly better
- vs DeepSeek V4 Pro+4.3, clearly better
- vs Claude Opus 4.8+6.1, clearly better
Claude Fable 5.1
- vs Claude Fable 5+1.3, inside the noise
- vs Gemini 3.8 Flash+3.4, inside the noise
- vs DeepSeek V4 Pro+4.3, inside the noise
- vs Claude Opus 4.8+6.1, clearly better
Claude Fable 5
- vs Gemini 3.8 Flash+2.1, inside the noise
- vs DeepSeek V4 Pro+3.0, inside the noise
- vs Claude Opus 4.8+4.8, clearly better
Gemini 3.8 Flash
- vs DeepSeek V4 Pro+0.9, inside the noise
- vs Claude Opus 4.8+2.7, inside the noise
DeepSeek V4 Pro
- vs Claude Opus 4.8+1.8, inside the noise
- Row clearly betterClearly better
- Inside the noise
- Row clearly behindClearly behind
Judges
The same pages through a second judge, GPT-5.6 Terra. The Index stays the Claude judge's score.
No sign the Claude judge favors Claude models. The Claude judge puts the Claude-built models 4.8 Index points behind the OpenAI-built ones. Terra puts them 4.3 Index points behind, on the same scale. The difference is -0.5 points (95% interval -4.0 to +3.2), so a lean of up to 4.0 points either way is not ruled out yet.
The two judges order the 14 models with a Kendall tau of 0.68 (95% interval 0.11 to 0.78). Across all 155 pages it is 0.57, and 0.36 within a brief, where the question is which build of the same page is better.
The widest split: Claude Sonnet 5.5 is fourth under the Claude judge and ninth under Terra.
| Model | Claude judge | Terra | Panel |
|---|---|---|---|
| GPT-6 Astra | 166.4 | 280.7 | 2+0.50 |
| Grok 4.7 | 266.1 | 184.5 | 1+0.61 |
| GPT-6.1 Sol | 365.4 | 379.8 | 3+0.42 |
| Claude Sonnet 5.5 | 461.8 | 974.5 | 7+0.06 |
| GPT-6 Luna | 461.8 | 578.2 | 4+0.19 |
| Gemini 4 Argon | 661.6 | 776.1 | 5+0.11 |
| Muse Wonder | 760.5 | 874.8 | 8+0.01 |
| Claude Opus 5.5 | 860.4 | 1170.7 | 10-0.14 |
| GPT-5.6 Sol | 958.9 | 479.3 | 6+0.08 |
| Claude Fable 5.1 | 958.9 | 677.1 | 8+0.01 |
| Claude Fable 5 | 1157.6 | 974.5 | 10-0.14 |
| Gemini 3.8 Flash | 1255.5 | 1369.6 | 12-0.41 |
| DeepSeek V4 Pro | 1354.6 | 1269.8 | 13-0.45 |
| Claude Opus 4.8 | 1452.9 | 1469.5 | 14-0.55 |
- Order is by each judge's point score, 1 for the highest. Each judge's score is its own Index on the same formula. Panel is the mean of the two judges' grade sums, each standardized across the 155 pages, in standard deviations.
- Terra grades higher overall (mean grade sum 28.1 against 23.8), so each judge's Claude-minus-OpenAI gap is measured in its own spread and shown in Claude Index points. Intervals: bootstrap, B=4000, resampling briefs, builds and passes. OpenAI Decisions has not judged model pages yet (v3 judges-2026-10-07 ran on websites and screenshots only).
Profile
The seven grades behind the Index, on the 1–5 scale (axis zoomed to the range used), mean of desktop and phone. Pick up to four models.
Rams Bench
The first track of the Index. Single-file pages built from written briefs, rendered and judged as a designer would see them.
5 public briefs in season 0.5. The private season-1 set (25 briefs plus these 5 as anchors) runs in v4.
Hierarchy, typography, spacing, color, components, composure, craft. Each graded 1–5 at desktop and at phone size.
How closely two judge passes on the same page agree, on 15 pages judged twice. Visual score 0.75, Terra grades 0.91.
Resamples briefs, builds within a brief and passes within a page, paired across models.
LandingGPT-6 Astra · 72.3 DashboardGPT-6.1 Sol · 68.8 SettingsSonnet 5.5 · 56.2 PricingGrok 4.7 · 68.8 ArticleGPT-6 Astra · 81.2
Method
Everything the number depends on, stated once.
What is the Rams Index?
A leaderboard of how well models design interfaces. Each model builds the same pages from the same written briefs. Rams renders every page at 1440 and 390 px and grades it on seven dimensions a designer would look at. The Index is the sum of those grades, rescaled to 0–100.
Why the grade sum and not one overall score?
Seven separate grades repeat better than a single verdict. On real webpages the grade sum separates design-led, typical and rough sites more clearly than our finding-based visual score, and two passes on the same page agree more often. The visual score is still shown, as a secondary column.
How is rank decided?
Rows are numbered by Rams Index, 1 for the highest. A model is clearly better than another when the paired 95% interval on the difference sits above zero. Each row also shows its rank range: every place those comparisons allow, from 1 plus the models clearly better to the total minus the models clearly worse. A wide range is a finding, not an omission: the board cannot order that model yet.
Does a Claude judge favor Claude models?
We run the same pages through GPT-5.6 Terra as a second judge and print it beside the score. No sign the Claude judge favors Claude models. The Claude judge puts the Claude-built models 4.8 Index points behind the OpenAI-built ones. Terra puts them 4.3 Index points behind, on the same scale. The difference is -0.5 points (95% interval -4.0 to +3.2), so a lean of up to 4.0 points either way is not ruled out yet. The Judges section of the Index shows each model's order under both. OpenAI Decisions (planned, v4) is the third judge from v4.
Who checks the judge?
Designers. This is season 0.5, published before their read: working designers rank the same pages blind, and the read rule was fixed before any of their answers were opened. Until that read (2026-11-06), this is the judge's opinion: stable and repeatable, not yet validated by designers.
What is private?
The brief texts of the season-1 set and the judge prompt. The method, the scores, the intervals and the sample pages are public. Five public briefs run alongside the private ones as anchors, so a model that overfits the public set shows up as a gap.
The full method, with every known weakness: How Rams measures design
Put the same eye on your own UI
The judge that graded these pages reviews the UI your agents write. Run it in Claude Code, Cursor or Codex while you build, or on every pull request.
Frameworks