Open-weight ecosystem

Local runtimes — Local runtime

Open-weight models on your own hardware, where the question is memory arithmetic rather than a price per token.

Local runtime Vendor page read August 15, 2026 benchr

What you can do with it

  • Keep every prompt and document on the machine, with no request leaving it.
  • Work offline, which is the whole argument in regulated and disconnected environments.
  • Trade a per-token bill for a fixed hardware ceiling.

In the reference layer

Models it runs

Where this leads