Offramp
Nebius Token FactoryNVIDIA Nemotron

Test which LLM calls can move to Nemotron.

Offramp finds every OpenAI and Anthropic call in a Python repo and shows, on held-out cases, which ones NVIDIA Nemotron on Nebius Token Factory can take over, by checking them against reference answers: seeded, computed by code, or written by another model family. It opens a pull request for the calls that reach the parity bar. The rest stay on their source model, with the numbers that show why.

Each call tries the cheapest tier first:

  1. Nemotron 3.5 Lightning
  2. Nemotron 3 Super
  3. Nemotron 3 Ultra
Watch demo run (replay)A recorded run. No backend or keys needed.

Two more recorded runs: the gate saying no (same app, Ultra not allowed) and a real open-source repo.

Try it on your repo

Paste a public GitHub repo and see every OpenAI and Anthropic call Offramp would test, with its list-price gap to Nemotron. No key, no sign-in.

Scan a repo

Or on your machine, same scan:

uvx --from git+https://github.com/vj-io/offramp offramp scan .

The recorded run, locally with no key: make demo. A full run measures parity with your own Token Factory key: uv run offramp run <repo>