JetBrains announced the release of Mellum2.1 on October 8, 2026. The model is offered under the Apache 2.0 license, meaning anyone can use or modify it without paying royalties. It is built as a mixture‑of‑experts (MoE) system with 12 billion total parameters, though only 2.5 billion are active at any given time.
How does Mellum2.1 perform?
The company says reinforcement learning on real code repositories raised its SWE‑bench Verified score—a benchmark for software‑engineering tasks—from 2.0 to 47.0. That jump suggests the model is much better at generating correct code snippets and fixing bugs than its predecessor.
What this means for developers
Because the model is open and free to use, developers can integrate it into their own IDE extensions or custom coding agents without waiting for a commercial API. The higher benchmark score should translate into fewer incorrect suggestions, saving time during coding sessions.
Why it matters
Developers can now access a powerful, open‑source coding assistant that is significantly more accurate on benchmark tests. The higher score should reduce the number of wrong code suggestions, making everyday coding smoother. Because the model is free to use, small teams and hobbyists can experiment without paying for proprietary APIs.
