JetBrains releases Mellum2.1, an Apache 2.0 12B coding model for agents

JetBrains says Mellum2.1 is a 12B mixture-of-experts model with 2.5B active parameters, trained with reinforcement learning in real environments and available now on Hugging Face under Apache 2.0. It claims about 1.6x faster single requests with multi-token prediction. GGUF builds and the vLLM component are coming soon. Speed and quality figures are JetBrains' own.

HEAT

OUTLETS

1

NEW IN 6H

+1

FIRST SEEN

Oct 11, 2026, 16:05 ET

LAST SIGNAL

40m ago