Lumma-Fev-4B

Use it to make structured decisions such as routing support tickets, classifying text, or scoring items against defined criteria.

Availability

Model weights available from Hugging Face; local inference and an API server are documented.

Pricing model

$0 self-hosted, according to the benchmark table.

Deployment

Local inference on CUDA, MPS, or CPU; can be served as an API.

Privacy

Not established

Licence

Apache-2.0

• The model returns a probability distribution for each typed question in a single forward pass. ^[1]

  It reads one document (the state) and a set of typed questions, and returns a probability distribution for each question in a single forward pass.

• The model never generates text. ^[1]

  It never generates text, so there is nothing to parse and nothing to hallucinate.

• The model can be loaded with Transformers or the lumma-fev package. ^[1]

  model = AutoModel.from_pretrained("FrontiersMind/Lumma-fev-4b", trust_remote_code=True,dtype=torch.bfloat16) # the tokenizer is loaded with it

• The lumma-fev package supports local inference on CUDA, MPS, or CPU. ^[1]
• The lumma-fev-serve package exposes a TypeSafe POST /v1/systemone API contract. ^[1]

  `lumma-fev-serve` exposes the TypeSafe `POST /v1/systemone` contract, so existing TypeSafe clients work by changing

• The benchmark table lists self-hosted use at $0. ^[1]

• ElevenLabs introduces two text-to-speech models Daily brief 2026-09-28