FrontierThe story, in brief
JetBrains Releases Mellum2.1: A 12B MoE Open Model for Coding Agents
47-point jump. JetBrains' 12B MoE model went from 2.0 to 47.0 on SWE-bench Verified with RL in real codebases.
The KeyNews take
Why it matters
JetBrains released an open-weight coding model with substantial SWE-bench gains through repository-grounded RL, establishing a practitioner-accessible baseline for agentic coding workflows and competing directly with closed frontier models in the coding domain.
The key facts
5 to knowMellum2.1: 12B total parameters, 2.5B active (MoE architecture)
SWE-bench Verified: 2.0 → 47.0 (23.5x improvement)
Apache 2.0 license (open-weight, commercially usable)
RL training in real repositories (grounding method)
Purpose: coding agents
The story so far
Earlier coverage of this storyline
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: JetBrains released Mellum2.1, an Apache 2.0, 12B mixture-of-experts thinking model with 2.5B active parameters. RL in real repositories lifted its SWE-bench Verified score from 2.0 to 47.0.