JetBrains open-sources Mellum 2.1 Thinking: 12B MoE with 2.5B active params

JetBrains · hf · 2026-10-10

JetBrains released Mellum2.1-12B-A2.5B-Thinking, a thinking-capable MoE text-generation model fine-tuned from Mellum2-12B-A2.5B-Base, with 12B total and 2.5B active parameters. Apache 2.0 licensed, it is trending on Hugging Face.

Original post →

More from Models

Models channel →