Updated

Mellum2.1

JetBrains released Mellum2.1, an open 12B mixture-of-experts model for coding agents. Its post-training uses reinforcement learning in real coding environments, enabling it to explore repositories, edit files, and check its changes; the model is available on Hugging Face under Apache 2.0.

By No Code MBA · Facts checked

Provider
JetBrains
Announcement date
Status at last check
Generally available
Model family
Mellum
Preceding model
Mellum2 Thinking
Capabilities
Text, Coding

What changed

Mellum2.1 keeps the Mellum2 architecture (12B total, 2.5B active parameters, 64 experts with 8 active, 131,072-token context) and changes post-training, mostly reinforcement learning in coding environments. JetBrains reports SWE-bench Verified at 47.0, LiveCodeBench v6 at 82.0 versus 69.4 for Mellum2 Thinking, and BFCL v4 at 62.3. The release includes Thinking, Instruct and Base checkpoints.

Access and availability

Weights are on Hugging Face at JetBrains/Mellum2.1-12B-A2.5B-Thinking under Apache 2.0. JetBrains documents serving with vLLM, SGLang, Transformers and Docker Model Runner.

Official access documentation

What you can build

Try it on a representative codebase task that includes diagnosing a failing test, proposing a patch, and checking the result; compare quality and latency with your current coding model.

Limitations

JetBrains' own table shows Qwen3.5 9B ahead on SWE-bench Verified, SWE-bench Pro, Terminal-Bench 2.1 and GPQA Diamond. GGUF builds and the MTP speculative decoding head were listed as coming soon, and no inference provider hosts it.

Sources and release history