# Sber Open-Sources a 432-Billion-Parameter Reasoning Model Under MIT License

GigaChat 3.5 Reasoning activates 28 billion parameters per token and, Sber says, uses 37 percent fewer reasoning tokens than DeepSeek V4 Flash Preview on competition math sets.

- Published: 2026-09-12T06:23:24.914Z
- Canonical: https://polylog.news/ai/2026-09-12/sber-open-sources-a-432-billion-parameter-reasoning-model-un
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [Hugging Face](https://huggingface.co/ai-sage/GigaChat3.5-432B-A28B-Reasoning), [Benzinga](https://www.benzinga.com/content/60416635/sber-launches-gigachat-3-5-ultra-advance-ai-development-enhanced-coding-long-context-processing)

Sber has released GigaChat 3.5 Reasoning, a mixture-of-experts model with 432 billion total parameters and 28 billion active per token, with weights published on Hugging Face and on the Russian code host GitVerse under the MIT license. The…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-09-12/sber-open-sources-a-432-billion-parameter-reasoning-model-un (subscription information: https://polylog.news/pricing).