JetBrains Mellum2.1: Fast Open Coding Agent Model
JetBrains has released Mellum2.1, an open model built for coding agents. The company says it is fast and ready for real agentic work—not just autocomplete. That matters because teams want models they can run and tune themselves without paying per token.
What is Mellum2.1?
It is the latest version of JetBrains' open model for AI-assisted software development. The blog post frames it as a model that "gets to work" inside agent loops: quick turnarounds, tool use, and fewer stalls. JetBrains is positioning Mellum2.1 against other open coding models, though the announcement does not name rivals or release benchmark numbers.
Why does a faster coding model matter?
Coding agents spend most of their time waiting. Faster inference means more iterations on the same task: edit, run tests, read errors, fix, repeat. Open weights also let companies host the model internally, which is a real draw for teams that cannot send code to third-party APIs.
What does it mean for JetBrains?
This looks like JetBrains trying to own more of the agent layer inside its IDEs. If Mellum2.1 performs well, the next step is likely tighter integration with IntelliJ and other JetBrains tools. The obvious competitor is code-specific models from OpenAI, Anthropic, and smaller open model labs. JetBrains has an advantage: it already lives where developers work.
No pricing or context-window specs were in the announcement, and no independent benchmarks were attached. That makes the speed claims worth testing rather than trusting.