A full-fledged neural network capable of operating without connection to ground servers has been launched on board the International Space Station for the first time. This was announced by the commander of the orbital laboratory, Sergei Kud-Sverchkov.
We are talking about the GigaChat model from Sber, which was delivered to orbit in November 2025 on the Soyuz MS-28 spacecraft. The system functions completely autonomously: it processes the crew's speech, converts it into text, and generates reports, notes, or brief summaries without delays in data transmission to and from Earth.
Kud-Sverchkov explained that the neural network is currently being trained to recognize human speech in the real conditions of the station. The AI is already keeping protocols of communication sessions, preparing messages for ground services, and structuring work records. Cosmonaut Sergei Mikaev is also participating in the testing.
Particular attention was paid during development to adaptation to the specifics of the ISS. The algorithms are configured to work in conditions of constant background noise, recognize professional abbreviations, technical terms, and simultaneously process Russian and English speech. This allows the crew to use the assistant in everyday tasks without additional settings.
The autonomy of the model is a key advantage for space missions. In case of communication delays or temporary absence of a channel with Earth, the neural network continues to work, ensuring the continuity of document flow and data recording.
Meanwhile, the operation of the ISS may last longer than planned: the American side has received funding to support the station's operation until 2032. This gives time for full-fledged testing of autonomous neural network solutions in long-term space missions.