Neptune 27B is in training
CompanyAIN / 01

An AI-native company for agent-native systems.

Ainfera builds open-weight models for systems that use tools, preserve state, and recover across long-running tasks. The company is organized around one durable output: a model teams can own, operate, and judge on explicit evidence.

Neptune 27B model card
01Company

The company is organized around the model.

AI-native means the model is not an add-on to a conventional product. Research, evaluation, interface design, and publication all meet at the release itself. Neptune is Ainfera's first model family and the reference point for that company architecture.

01Object
Open-weight models for agentic systems
02Method
System-level evaluation with explicit conditions
03Release
Weights, interface, limits, and evidence record
02Factory

The factory is AI-native, too.

The factory treats model definition, agent interface, evaluation, and release record as one connected system. The work is not finished when a weight file exists. It is finished when the intended behavior and the conditions of every claim are legible.

AI-native factory registerA four-stage register connecting model definition, interface, evaluation, and release record.FACTORY REGISTER / ONE RELEASE OBJECTDEFINITION → EVIDENCE → RECORD01020304DEFINEINTERFACEEVALUATERELEASE
01 / Define

Name the operating envelope.

Identity, interface, context, and deployment intent begin in the same record.

02 / Evaluate

Test the complete system.

Tools, state, recovery, hardware, and runtime remain attached to the observation.

03 / Release

Ship a model with its receipt.

Public numbers stay N/A until their method and conditions can travel with them.

03Founder
Hizrian Raz, founder of Ainfera
Solo founder

Hizrian Raz

Hizrian founded Ainfera to build model companies differently: start with a model worth owning, make the agent system part of its definition, and keep public claims inside the boundary of available evidence.

Built by agents, for agents.

Factory