JetBrains releases Mellum2.1, a 12B open model for coding agents

TL;DR

• On 8 October JetBrains released Mellum2.1, an Apache 2.0 mixture-of-experts model with 12B parameters, 2.5B of them active per token.
• Reinforcement learning in real repositories took its SWE-bench Verified score from 2.0 to 47.0 in JetBrains' own tests.
• Qwen3.5-9B still leads on all three agentic tests in JetBrains' table: SWE-bench Verified, SWE-bench Pro and Terminal-Bench 2.1.