Most AI comparison sites give you a number. This one gives you the work. Four frontier models built the same projects from identical prompts, and every result is recorded, playable and shown next to the others. Pick a brief below and watch all four at once; that is the whole comparison, and it takes about thirty seconds.
How do you compare AI models on this site?
Four ways, each one click from the stage above.
- All four at once. The default view plays every model's build of one brief in sync, so differences in motion and layout jump out.
- Two side by side. Pick any pair to compare closely, or open one of the six model pairs to see that pair on every brief.
- One at a time, live. Every build opens in a sandbox exactly as the model wrote it, so you can click, scroll and play.
- Blind. Names are hidden and the order is shuffled; you choose first, then reveal.
What exactly is being compared?
Claude Opus 5.5 and Claude Fable 5.1 running in Claude Code, and GPT-6 Astra and GPT-6.1 Sol running in OpenAI's Codex, all at maximum reasoning effort. Each brief fixes the content, so models cannot win by inventing a better product, and each model gets one attempt with a 150-minute cap. The methodology publishes the exact commands and recording scripts.
Why compare AI models visually instead of by score?
Because the things you care about do not fit in a score. A benchmark can tell you a model fixed 70% of known bugs; it cannot tell you whether its landing page looks designed or assembled, or whether its game is fun. Popular comparison sites mostly aggregate one of two things: crowd votes, as on LMArena, or benchmark scores, as on Artificial Analysis. Both are useful and both are abstractions. Here you see the artefact itself, alongside facts such as time, cost and errors.
What should you look for when you compare?
Start with the brief's checklist: every brief page lists what to watch for, such as whether a dashboard's numbers match its data or whether a physics simulation lets balls tunnel through walls. Then use the facts below each build. For the four models' prices and specifications in one table, see the AI model comparison chart.
The fairest comparison is the one where you cannot see the logos. Try it on four landing pages.
Take the blind test
Frequently asked questions
What is the best way to compare AI models?
Give them the same task, the same instructions and the same limits, then compare the outputs without knowing which model made which. That is what GlanceHall does for coding work, with the builds recorded and playable rather than reduced to a score.
How is this different from LMArena or benchmark leaderboards?
Arenas aggregate votes on short prompts and leaderboards publish scores on fixed tests. Here every model builds a complete project from an identical brief, and you see all the results at once. There is no vote count and no ranking, only what each model made.
Which AI models can I compare here?
Claude Opus 5.5 and Claude Fable 5.1 from Anthropic, and GPT-6 Astra and GPT-6.1 Sol from OpenAI, each running in its vendor's own coding agent at maximum reasoning effort. More vendors are planned for the next round.
Is the blind test saved or counted?
No. Your pick stays in your browser and is never sent to us or tallied. The blind test exists to protect your own judgement from brand bias, not to build a leaderboard.