Mellum2.1 targets coding agents with stronger repository training

JetBrains reported agent scores for Mellum2.1 and comparison models

Listen to this article

JetBrains publishes Mellum2.1 Thinking with 12 billion parameters and 2.5 billion active under Apache 2.0. Training uses real repositories with shell and file editing tools.

JetBrains reports 47 percent on SWE bench Verified, compared with 2 percent for Mellum2 and 50 percent for Qwen3.5 9B in its shared evaluation setup. These are vendor results, not independent testing.

Deployment details remain important. The card provides vLLM instructions but describes GGUF builds and the prediction head for speculative decoding as coming soon. It does not establish universal local hardware compatibility.

Read the JetBrains model card

Release date unverified.

Chart by ByteForward using JetBrains reported results

Jordan Reid

Jordan Reid is focused on AI tools, agents, developer products, and the way technology changes everyday work. Jordan approaches a launch from the user’s side of the screen. What can it actually help someone finish? The voice is practical, conversational, and skeptical of products that turn a simple job into five new settings. Coverage follows coding assistants, creative software, browser agents, and the workflows around them, with attention to pricing, permissions, setup, and the human work that remains.