The only build to clear every mandatory gate. Official score 89.25/100.
Intentional design · reliability first
Eight models. One identical website brief.
Every model received the same starter, prompt, harness, tool limit and evaluation. Live builds preserve the model output while applying only the disclosed path adaptations required to serve them together.
8 scored builds$6.45 total run cost1 qualified build7 not qualified
Category leaders
Different questions have different winners.
These badges describe the strongest result in each measured category. They do not replace the qualification rule or establish a universal model ranking.
Highest blind visual-review result at 19/20. Not qualified because mandatory gates failed.
Highest raw score at 91.75/100 and highest code score at 8.75/10. Not qualified because one mandatory gate failed.
Qualified · 1 build
Cleared the mandatory bar
Qualified results can receive their full official score.
xAI / SpaceXAIQualified
Reliability winner
Grok 4.5
Raw score
89.25/100
Official score 89.25/100 · uncapped
Auto 55/60Visual 17.5/20Code 6.75/10Verify + efficiency 10/10
Mandatory gatesAll passed
Not qualified · 7 builds
Raw performance, with failed gates disclosed
These entries are ordered by raw score for diagnosis, not presented as an official second-through-eighth ranking.
Z.ai / Zhipu AINot qualified
Raw + code winner
GLM-5.2
Raw score
91.75/100
Official score capped at 49/100
Auto 56/60Visual 17/20Code 8.75/10Verify + efficiency 10/10
Failed mandatory gate
F05Form errors and deterministic success
AnthropicNot qualified
Claude Fable 5
Raw score
83.25/100
Official score capped at 49/100
Auto 51/60Visual 14/20Code 8.25/10Verify + efficiency 10/10
Failed mandatory gate
F04Dialog focus trap, Escape and focus return
OpenAINot qualified
Visual winner
GPT-5.6 Sol
Raw score
77.25/100
Official score capped at 49/100
Auto 41/60Visual 19/20Code 7.25/10Verify + efficiency 10/10
Failed mandatory gates · 4
F04Dialog focus trap, Escape and focus returnF05Form errors and deterministic successR01-tabletNo horizontal overflow at 768pxA01No serious or critical axe violations
Moonshot AINot qualified
Kimi K2.7 Code
Raw score
74.75/100
Official score capped at 49/100
Auto 40/60Visual 17.75/20Code 8/10Verify + efficiency 9/10
Failed mandatory gates · 2
F04Dialog focus trap, Escape and focus returnA01No serious or critical axe violations
Xiaomi MiMoNot qualified
MiMo-V2.5-Pro
Raw score
66.5/100
Official score capped at 49/100
Auto 41/60Visual 10/20Code 5.5/10Verify + efficiency 10/10
Failed mandatory gate
F04Dialog focus trap, Escape and focus return
MetaNot qualified
Llama 4 Maverick
Raw score
39.25/100
Official score 39.25/100 · raw result already below the cap
Auto 26/60Visual 8.25/20Code 0/10Verify + efficiency 5/10
Failed mandatory gates · 5
F02AFiltering, URL state and direct loadingF03Keyboard-operable service detailF04Dialog focus trap, Escape and focus returnF05Form errors and deterministic successA02Page accessibility while modal is open
MiniMaxNot qualified
MiniMax-M3
Raw score
31.5/100
Official score 31.5/100 · raw result already below the cap
Auto 18/60Visual 3/20Code 3.5/10Verify + efficiency 7/10
Failed mandatory gates · 6
F02AFiltering, URL state and direct loadingF03Keyboard-operable service detailF04Dialog focus trap, Escape and focus returnF05Form errors and deterministic successA01No serious or critical axe violationsA02Page accessibility while modal is open
Visual and code scores were produced by two independent blind AI reviewer agents. Neither received model identities or the other reviewer’s scores. This is a limitation of Round 0 and must remain disclosed when results are cited.