Calls tools without translation.
Prompts, schemas, arguments, observations, and results remain explicit across the exchange.
Ainfera is building Neptune 27B, an open-weight model for agentic systems. It is designed for native tool use, persistent state, long-horizon execution, and recovery.
Neptune 27B is an open dense, approximately 27-billion-parameter model for agent systems. Its product definition joins model identity, agent-native behavior, deployment intent, and an evaluation record.
Public benchmark data for Neptune 27B is N/A. When results are available, every value will carry the exact model, suite, runtime, date, and uncertainty needed to interpret it.
A useful agent must keep intent, state, tool outputs, and recovery connected. Neptune treats the task as a continuous system, not a stack of isolated replies.
Prompts, schemas, arguments, observations, and results remain explicit across the exchange.
Plans and state transitions remain connected to the original task across multiple steps.
Rejected calls and failed paths stay inspectable, so the agent can correct and continue.