Why Dropstone runs open-weight models
The models behind Dropstone Fast, Pro, and Heavy are open-weight, so the weights are published and anyone can check which model is behind a tier. Here is what that gives you, and what it does not mean.
The models behind Fast, Pro, and Heavy are open-weight. Their weights are published by the people who trained them, rather than kept behind a private endpoint.
Open weights make the lineup checkable. You do not have to take a claim about which model is behind a tier on trust, because the model itself is published and can be run by anyone.
What open weights give you
| What it gives you | Why it matters |
|---|---|
| The claim is verifiable | When Dropstone says a tier runs a particular model, that model is a published artifact with a public name, not an internal build you cannot look up |
| The evaluation can be repeated | The tests behind a pick are run against models anyone can obtain, so the result is something another party could reproduce rather than a private benchmark |
| No tier is locked to one lab | If a better open-weight model appears, it can take a slot. A tier's future does not depend on one company's roadmap or one company's pricing |
Why open weights rather than closed models
A closed model is a black box with a name attached, and it can be changed, repriced, or withdrawn without telling you. Open weights behave differently: the artifact is fixed and named, so a tier can be pinned to a version and a change is a visible event rather than a silent one.
That is the property the lineup needs. A product that promises you always know what is running cannot be built on models that change underneath you without saying so.
What open weights do not mean
| What it does not mean | The plain answer |
|---|---|
| That Dropstone is open source | It is not. Open weights describe the models, not the product, and Dropstone itself is a commercial service |
| That you have to run anything | You do not. You can run these models yourself if you want to, but Dropstone exists so you do not have to: no servers to rent, no serving stack to build, and no evaluation cycle of your own |
| That every model is available | It is not. A model is promoted to a tier only if it passes the evaluation, safety included. How Dropstone chooses its models covers that |
Which models are open-weight?
The lineup is on Which AI models does Dropstone use?, with the maker of each model.
Related articles
- Which AI models does Dropstone use?Dropstone runs three tiers. On the 1.8 lineup, Fast runs DeepSeek V4.1 Flash, Pro runs GLM-5.3 Flash, and Heavy runs Kimi K3, re-evaluated every month.
- How Dropstone chooses its modelsEvery month Blankline re-evaluates the open-weight models available for each Dropstone tier against three tests, how well it does the work, what it costs to serve, and how safely it behaves with tools. A model has to beat the one already in the slot.
- Model versions and the monthly lineupDropstone numbers each lineup with the year and the month it was cut, and a new version ships only when a tier's pick changes. How to read the number, and which lineup you get.
- Choose a modelHow to pick between Dropstone Fast, Pro, and Heavy, where the selector lives in chat and in the CLI, and why the name Fast describes what a tier costs rather than how fast the model inside it runs.
- Where your requests runDropstone's servers and the inference providers that run its models are hosted in the United States. Requests are never routed through providers outside the US. What that means for your data.