GPT 6.1 Sol
gpt-6.1-sol · mediumThe 30 September v1 cohort uses separate fresh agents for skills off and on, with all five original briefs in each condition. The skill condition uses the pinned Anthropic frontend-design snapshot.
Read the run notesSame brief.
Different
instincts.
Compare websites built from five briefs by Sol, Luna, Astra, Claude and Space Bunny. Switch design skills on or off, then explore the results side by side.
· GPT 6.1 Sol joins v1 with five briefs, skills off and on. See the run notes.
Pick a brief. See what changes.
About the experiment.
A shared starting point
Each model receives the same five business briefs and makes its own decisions about layout, content, and interactions. Read the original brief on any card to see what was asked.
Two conditions, more context
Each brief has a version with and without the design skill. The GPT 6.1 Sol cohort and replacement Sol, Luna, and Astra runs use medium reasoning, with separate fresh agents for the baseline and skill conditions. These are exploratory comparisons, with no scores or rankings.
For a closer look, read the Borrowed Weekend comparison.
About the models
The AI models behind these websites, and how to read their labels.
GPT 6 Luna
gpt-6-lunaA GPT model in this collection. Luna builds websites from the same briefs as Sol and Astra, with separate versions for each skill condition.
Read the run notesGPT 6 Sol
gpt-6-solAnother GPT model used to turn the briefs into working websites. The replacement v1 run uses medium reasoning for Sol.
Read the run notesGPT 6 Astra
gpt-6-astraAnother GPT model in the collection. The replacement v1 run uses medium reasoning for Astra.
Read the run notesClaude Opus 5.5
claude-opus-5-5 · mediumBuilt all five briefs without skills first, then read the pinned Anthropic frontend-design skill and built them again in the same session.
Read the run notesClaude Sonnet 5.5
claude-sonnet-5-5 · mediumRan in its own fresh session: all five briefs without skills first, then the pinned Anthropic frontend-design skill in the same session.
Read the run notesSpace Bunny
space-bunny · alias · OpenCode harness · High requested · two v1 cohorts29 September 2026: A new paired run is available for all five briefs, both without a skill and with the pinned design skill. The earlier cohort remains available under From the archive and in the comparison viewer’s Legacy models menu. The earlier skill-enabled pages used different skill-file bytes, so differences between cohorts cannot be attributed to a model update alone. Read the new run notes
Space Bunny runs through OpenCode under an undisclosed alias. High reasoning was requested, but the harness did not expose a setting that confirms it. Its underlying identity remains undisclosed.
Rumored identity: MiniMax M3.1. This is unconfirmed community speculation. Rumor discussion
What is a stealth model?
A model presented under an alias while its underlying identity is undisclosed. An alias alone does not establish its provider, model family, or release version. The harness (such as OpenCode or Cline) is the tool running the model, not its provider. Treat stealth entries as the outputs of that named run; their identity remains unknown until disclosed.
What does reasoning mean?
Labels such as low, medium, high, and max record the reasoning setting used for a run. They are settings, not rankings. The replacement v1 runs use medium for Luna, Sol, and Astra; other cohorts may use different settings.
What does “skills on” mean?
The model received additional design instructions. It is the same model with extra guidance. Different runs may use different instructions.
Names on this site identify the runs collected here; they do not establish public availability or a confirmed identity behind an alias. Earlier Space Bunny, GPT 5.6 Sol, GPT 5.6 Luna, Claude Opus 4.8 and Claude Opus 5 builds live in the archive.