Modelglass — AI model & pricing registry
Know what your AI costs before you commit.
New models launch weekly and prices shift constantly. Modelglass is the layer that decides for you: it sits in front of your prompt, knows what every model costs and what each one is good for, and picks the lowest-cost one that can do the task. Read the data yourself, or wire it in and let it choose.
174 models · 56 providers · 5 model types · updated daily
The origin
The cost of not knowing what you're creating
AI is everywhere now — it's not unusual to be paying for more than one model or tool at a time. There's real overlap between them, and real differences too, but those differences don't mean the same thing to every person or business. The landscape is filling up fast, it's hard to track what's changed, and even harder to justify switching once you've settled somewhere.
Modelglass was built with exactly that uncertainty in mind — by someone who can't leave a set of overlapping options unsorted, and would rather maintain the comparison once than rebuild it by hand every few weeks.
It sits in front of your prompt, not behind it. It tracks what every model costs and what each one is good for — updated daily — and turns that into a single answer: the lowest-cost model that can actually do the job.
Read it like a reference, or wire it into your own workflow and let it do the choosing. Either way, you stay free to move between models as they change. See how our routing works →
We don't collect your data or your prompts. We're a pricing and capability registry, and we'd love to hear from you.
In Andreessen Horowitz's 2025 survey of 100 enterprise CIOs, 37% now run five or more AI models in production — up from 29% a year earlier.
One product, three ways it earns its place
The same registry, doing three jobs
Modelglass isn't three tools. It's one dataset — model pricing and capability, kept current — that works whether a person is reading it or a system is calling it.
- 01
It informs the user
On its own, the registry answers one question well: which model is right for this task, and what it costs. People use it to choose directly; teams wire it into their own product so their customers get the same answer.
Browse the registry - 02
It picks the lowest-cost model for the task
Wired into a router, the same data selects the cheapest model that still clears the bar for the job — and shows the cost before it spends it. Bring your own keys; nothing routes through us.
How routing works - 03
It keeps businesses better informed
Prices and specs move constantly. The registry and its news feed update daily, so a price cut or a new release reaches you early — not on your next invoice.
See the news feed
Questions people ask
A fast-moving, fragmented market — in plain terms
- Everyone's paying for a few different AI tools now. How do I tell what's actually different between them?
- Start with the two things that vary most: what each one costs per unit of work, and what it's actually good at. Those are the columns the registry keeps current. The differences are often smaller than the marketing suggests — and occasionally far larger than the price gap implies.
- A new model lands every couple of weeks. Do I need to care about all of them?
- No — most aren't a reason to change anything. The ones that matter either beat what you're using on a task you actually run, or undercut it on price. That's a short list, and it's the one the news feed is built to surface.
- We've built our product around one model's API. What's the real risk in switching?
- The risk is rarely the integration — it's not knowing whether a new model behaves the same on your inputs, and not having a way back if it doesn't. Knowing the cost and capability gap up front turns “should we switch” from a leap into a measured call.
- Where does your pricing data come from?
- Published provider pricing, checked against each provider's own pages and updated daily. Every figure carries its source. When a price can't be verified, it's marked rather than guessed.
- Does Modelglass see my prompts, my outputs, or my usage?
- No. Nothing you send to a model passes through us. Modelglass is a registry you read — or that your own systems query — not a proxy in the request path. The routing tooling runs on your keys, on your side.
- Is this something I use, or something I build on?
- Both. The site is a free comparison tool anyone can use. The same data is available as an API and an MCP feed for teams wiring model choice into their own routing, pricing pages, or agents.