JetBrains releases Mellum2.1, a 12B open coding-agent model under Apache 2.0
Overview
JetBrains has released Mellum2.1, a 12B mixture-of-experts thinking model with 2.5B active parameters, under the Apache 2.0 license on Hugging Face, aimed at coding agents and fast sub-agents.
The released checkpoint is Mellum2.1-12B-A2.5B-Thinking. JetBrains says post-training shifted to reinforcement learning across thousands of environments and millions of sandboxed runs, and reports gains over Mellum2 on several benchmarks, including SWE-bench Verified.
The SWE-bench Verified score rose from 2.0 to 47.0 after reinforcement learning in real software repositories, according to JetBrains' self-reported results. Coverage notes that Qwen3.5-9B still leads on SWE-bench Pro, GPQA Diamond and AIME. JetBrains also says the model serves almost twice the tokens of Qwen3.5-9B under heavy load, a claim that is self-reported. GGUF builds for llama.cpp, Ollama and LM Studio, which start at 7.0 GB for local use, are described as coming soon.
Written by AI from the articles below · updated Oct 8, 7:52 PM ET
Check the sources:
Article timeline
Follow the coverage from different perspectives. Times are ET.
- MarkTechPostJetBrains releases Mellum2.1, a 12B MoE open model for coding agents
AIJetBrains has released Mellum2.1, a 12B mixture-of-experts thinking model with 2.5B active parameters, under Apache 2.0 on Hugging Face. Post-training reinforcement learning in real software repositories raised SWE-bench Verified from 2.0 to 47.0, according to JetBrains' self-reported results. Qwen3.5-9B still leads on SWE-bench Pro, GPQA Diamond and AIME, and GGUF builds start at 7.0 GB for local use.
- JetBrains AI BlogPickJetBrains releases Mellum2.1, an open coding model trained with reinforcement learning
AIJetBrains released Mellum2.1, a 12B mixture-of-experts model with 2.5B active parameters under the Apache 2.0 license, built for coding agents. Post-training shifted to reinforcement learning across thousands of environments and millions of sandboxed runs, and the model is available on Hugging Face. The source reports gains over Mellum2 on LiveCodeBench, AIME, GPQA Diamond, BFCL v4, IFEval, and SWE-bench Verified, and says it serves almost twice the tokens of Qwen3.5-9B under heavy load.
Heat trend
Not enough continuous observations to show a trend yet.