On 27 July Anthropic put out a statement titled "Our position on open-weights models". It is a policy post rather than a product launch, and it will not change what Claude does tomorrow.
It is worth reading anyway, because the question underneath it — who holds the weights of the model your team depends on — decides how much of your workflow you actually control.
Why a policy post matters to a tools decision
Every assistant in this directory sits on top of somebody else's model. Cursor routes between vendors. GitHub Copilot now offers OpenAI, Google, Anthropic and xAI models inside one subscription. Emergent and Lovable call hosted APIs you never see.
That arrangement is convenient until a model is deprecated, repriced or withdrawn. All three happened during the fortnight this article covers.
The practical read
If a vendor never releases weights, a model can disappear from your toolchain on the vendor's schedule and there is no fallback you can run yourself. If weights are released, a capable model stays available to you indefinitely, at the cost of running it.
For most teams this is not an argument for self-hosting. It is an argument for not building a critical process on a single model name, and for checking whether your assistant lets you switch.
- Copilot and Cursor both let you change model — check the setting exists before you need it
- Prompt-built app tools rarely expose model choice at all
- n8n is the one automation platform here you can host yourself
What it does not say
The post is a position, not a commitment to release anything. Nothing about Claude's availability changes as a result of it, and no Anthropic model became open-weight on 27 July. Treat it as context for procurement conversations rather than news you need to act on this week.