Mellum2.1 Gets to Work: A Fast Open Model for Coding Agents

Summary

JetBrains open-sourced Mellum2.1, a compact 12B mixture-of-experts model with only 2.5B active parameters, released under the Apache 2.0 license. The architecture is unchanged from Mellum2, so all improvements come from post-training through reinforcement learning now the main training phase. The model trained on millions of sandboxed runs across thousands of real environments to develop agentic coding skills like exploring codebases, editing files, and checking its own changes. Performance evaluations show substantial gains in agentic coding, general coding, competitive programming, math, tool calling, and general knowledge compared to Mellum2 and similar open models like Qwen3.5-9B and Gemma 4 E4B. Mellum2.1 matches Mellum2’s speed and with multi-token prediction serves almost twice as many tokens as Qwen3.5-9B under heavy load.

Takeaways

1. 1

   Mellum2.1 uses multi-token prediction to speed up generation: about 1.6x for single requests and nearly twice the token throughput of Qwen3.5-9B under load.

2. 2

   The 12B MoE model has only 2.5B active parameters and is released under Apache 2.0, enabling private, self-hosted deployment.

3. 3

   Compared to Mellum2, the largest performance gains are in agentic coding, with improvements across coding, competitive programming, math, tool calling, and general knowledge.

FAQ