JetBrains has open-sourced Mellum 2, its latest large language model. The move makes the 12-billion-parameter model freely available under the Apache 2.0 license. Developers can now use it for code generation, editing, tool calling, agentic workflows, and reasoning tasks.
Sparse MoE architecture activates only 2.5B parameters per token.
Mellum 2 is a sparse Mixture-of-Experts model. It activates only 2.5 billion parameters per token, which keeps inference fast on standard hardware. The model comes in three versions: base, instruct, and thinking.
The context window has expanded to 131,072 tokens, up from 8,192 in the original Mellum. This allows the model to handle much larger codebases or documents in a single pass. JetBrains says the architecture is designed for efficient computation.
DataSpell discontinued, features consolidated into PyCharm Pro.
Alongside the model release, JetBrains announced it is discontinuing DataSpell. The company is consolidating DataSpell's features into PyCharm Pro. Users of DataSpell will see those capabilities integrated into the professional Python IDE.
Mellum 2 is available globally starting June 2026. The Apache 2.0 license permits commercial use, modification, and redistribution. Developers can download the model from JetBrains' official channels.
The original Mellum model launched with a smaller context window and fewer parameters. Mellum 2 represents a significant upgrade in both scale and capability. JetBrains continues to invest in AI tools for developers.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.