AgentJev-0.6B

Developers can use it as a gate, router, or scorer in an agent or workflow loop.

Availability

Weights available on Hugging Face

Pricing model

Not established

Deployment

Run locally; README documents an HTTP server

Privacy

Not established

Licence

Apache-2.0

• The checkpoint reads the task, code, and checks shown, then returns a probability that the work is finished. ^[1]

  It reads the task, the code, and the checks that were shown, then returns a probability that the work is actually finished.

• The model.safetensors file is the full module, including the backbone and candidate head. ^[1]

  It is the full module: backbone plus candidate head.

• The server binds to 127.0.0.1. ^[1]

  The server wants a torch checkpoint, so wrap the file once.

• The checkpoint is FP32 and uses Qwen3-0.6B as its backbone. ^[1]
• The model supports Boolean, choice, and score primitives. ^[2]
• Each primitive returns a full distribution over the supplied options or levels. ^[2]

  | Output | Text you hope is JSON | A distribution over options you supplied |

• The model decodes zero tokens. ^[2]

  The usual answer is to ask a 27B–70B model to talk its way to a boolean, then parse the talk.

• Weights are available on Hugging Face. ^[2]

  <a href="https://huggingface.co/aimeigaoshou/agent-jev"><img alt="Weights on Hugging Face" src="https://img.shields.io/badge/weights-Hugging_Face-C6A36A?style=flat-square"></a>

• The README provides instructions for running an HTTP server.
• The model has a 2,048-token context and refuses over-length input. ^[2]

  **The context is 2,048 tokens, and over-length input is refused.** A diff or a trace that does not fit is an error, not a silent crop of the question or a candidate.

• New models and tools focus on typed decisions and long-context tasks Weekly brief 2026-W39