The Model Atlas
Pick what you actually care about — quality, cheapness, speed, openness — and the atlas ranks a landscape of model archetypes by your priorities, not a vendor's leaderboard. The lesson lives in the reshuffle: the moment you weight anything other than raw capability, a different model jumps to #1, and the "highest-quality model" is rarely the right pick. It's a hands-on way to feel why "just learn tool X" ages so fast — the map moves under you. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Drag the four priority sliders — Quality, Cheapness, Speed, Openness — to describe what matters for your use case, or tap a preset (Quality first, Cheapest, Fastest, Open only, Balanced).
- Watch the ranked atlas re-sort live. Each model shows its weighted score and whether it ships open weights or is closed.
- Open the X-ray panel to see why it reshuffled: your #1 pick placed side by side with the raw quality leader, axis for axis.
- Copy or share your ranking — the gap between "best model" and "best for me" is the whole point.
What this clears up (the fundamentals)
- There is no single "best model", only a best model for a set of weights. The leaderboard is a function of your priorities; change them and it moves.
- Cheapness is inverted capability. A high price is a cost, so the atlas scores
(1 − cost)— a cheap model earns points exactly where an expensive one loses them. - Open vs closed is its own axis. Portability and open weights can outweigh a few quality points when lock-in, privacy, or on-prem matters — so an "Open only" run reshuffles the board hard.
- Churn is the signal, not the noise. The reshuffle you see when you nudge one slider is the same reshuffle the market shows month to month. Learning to read the map beats memorizing today's leader.
Where it's used
A top-of-funnel AI-engineering literacy check — the fastest way to feel why model choice is a tradeoff, not a ranking. It's a miatz build-lab concept playable: play it here, then learn to build the weighted-scoring rubric and the reshuffle reveal yourself in AIE-130.
FAQ
How is a model's score calculated?
Each model carries four 0–1 axes (cost, quality, speed, openness). Your slider weights (also 0–1) combine them as (1 − cost)·wCost + quality·wQuality + speed·wSpeed + openness·wOpenness, rounded to three decimals. Cost is inverted so cheaper scores higher; the models are then sorted by that weighted score. It's a deterministic rubric — no model call, no randomness.
Why isn't the highest-quality model always #1?
Because #1 is decided by your weights, not by quality alone. Give any points to cheapness, speed, or openness and a model that's merely good — but cheap, fast, or open — can overtake the capability leader. The "reshuffled" flag lights up exactly when that happens.
Are these real, named models?
No — they're evergreen archetypes (Frontier Flagship, Open Heavyweight, Budget Speedster, and so on) chosen so the tradeoffs stay true even as specific products change. The atlas teaches the shape of the decision, not this month's leaderboard.
What does "openness" mean here?
How portable and self-hostable a model is — open weights and a permissive license versus a closed, API-only offering. Models scoring 0.5 or above are tagged open weights; the axis lets you price lock-in avoidance into the ranking.
Is anything uploaded?
No. Your priorities are scored entirely in your browser — nothing is transmitted, stored, or logged. Turn off your Wi-Fi and it still works.
Limits
A teaching rubric with a fixed set of archetypes and four hand-set axes — a fast, opinionated way to internalize the tradeoff, not a procurement benchmark. Real model selection also weighs context window, tool-calling reliability, safety, data residency, and your exact workload. The move it teaches — score models against your weights instead of chasing a single leaderboard — is exactly the real practice.
Related
Part of the miatz playgrounds. Explore the rest from the playground catalog.
Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.