The Model Landscape Board
A no-hype buyer's guide to language models. Instead of a leaderboard that crowns one "best" model, this board holds a fixed set of models — priced in and out, sized by context window, tagged open or closed weights, and rated on quality and speed — and lets you set your real constraints: a price ceiling, an open-weights requirement, a minimum context window, a quality floor. It marks which models fit every bar and names the cheapest one that clears them all. The reveal is the honest part: "best" is never absolute — it's the cheapest model that meets your filters, and change one filter and a different model wins. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Load a use-case preset — Cheap chatbot, Coding agent, Open weights only, Long-context batch — or set the four filters by hand.
- Drag the max blended price, tick open weights only, pick a minimum context window, and set a quality floor.
- Read the board: fitting models are highlighted, misses are dimmed, and the cheapest fit is tagged.
- Open the X-ray to see why that model won the slot — and why each other model missed. Export the shortlist to compare later.
What this clears up (the fundamentals)
- "Best model" is a filter result, not a fact — the winner is the cheapest model that clears every bar you set. Move one bar and a different model wins. That's model selection, demystified.
- Blended price is the number to compare — input and output are priced separately, so a single blended
$/1M(the mean of the two) is the fair, one-glance comparison. - Open vs closed weights is a real axis, not a badge — open weights mean you can self-host or swap providers; closed means API-only. It changes your lock-in, not just your bill.
- Context window and quality tier trade against price — the biggest context and the top quality tier cost the most. Requiring both narrows the board fast; that narrowing is the point.
Where it's used
A FREE, top-of-funnel AI-literacy check — the fastest way to feel why one model got picked over another, as a filter you can watch, not a vendor's black box. It's a miatz build-lab concept playable: the same routing table an inference layer uses internally to decide "why did my run land on this model" is the public artifact here. Play it, then learn to build the filter and the structural verdict yourself.
FAQ
How is "blended price" calculated?
It's the mean of input and output price per million tokens: round((priceIn + priceOut) / 2, 2). One number lets you rank models at a glance instead of juggling two columns. Real bills depend on your input/output ratio, but blended price is the honest first cut.
Why does the "best" model change when I change a filter?
Because there is no absolute best — only the cheapest model that clears the bars you set. Raise the quality floor and a pricier model wins; require open weights and every closed model drops out. The board makes that dependency visible.
What does open vs closed weights actually mean for me?
Open-weights models can be downloaded, self-hosted, and swapped between providers, so you're not locked to one vendor. Closed-weights models are API-only. It's a portability and lock-in axis, not a quality claim — some of each are strong.
Are these live, real-time prices?
No. This board uses a fixed, illustrative set of models with round numbers chosen so the math is hand-checkable — it teaches the method of comparison, not this week's price sheet. The full miatz build-lab version wires the same filter to live, cached ranking feeds.
Is anything uploaded?
No. Your filters are applied entirely in your browser — nothing is transmitted, stored, or logged. Turn off your Wi-Fi and it still works.
Limits
A teaching board with a fixed roster and illustrative prices, not a live procurement feed. Blended price averages input and output equally, which flatters output-heavy workloads; quality and speed are coarse 1–3 tiers, not benchmark scores. The method it teaches — set your real constraints, compare on a blended price, and treat "best" as the cheapest model that fits — is exactly how a real routing layer decides.
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.