Mellum2.1

Mellum2.1 is an update to JetBrains’ Mellum2 coding model, published in October 2026 with open weights under the Apache 2.0 licence. It is a mixture-of-experts model with 12 billion parameters in total and 2.5 billion active, and its model card gives a context window of 131,072 tokens. JetBrains aims it at agentic coding work such as operating inside a repository, running commands and calling tools, and also at hard coding, math and reasoning problems without an agent setup. The announcement says multi-token prediction makes a single request about 1.6 times faster, and that on one H200 GPU under heavy load it serves almost twice as many tokens as Qwen3.5-9B. JetBrains compares it with Mellum2, Qwen3.5-9B and Gemma 4 E4B in charts that name LiveCodeBench, AIME, GPQA, BFCL, IFEval and SWE-Verified, but these are the vendor’s own runs and the text gives no numbers. The weights are on Hugging Face, with GGUF builds for llama.cpp, Ollama and LM Studio and a head for vLLM listed as coming soon. No hosted inference provider serves it yet, so teams would run it on their own hardware. No price is stated because there is no hosted API.

Not yet independently verified: Read on JetBrains’ own announcement and its Hugging Face model card. The announcement page shows no day, so the date of 8 October 2026 comes from its first appearance on Hacker News that day, and the card states no release date either. The speed and quality comparisons are JetBrains’ own, shown only as charts, so no score is stored. We will update this when it can be confirmed, and remove this note.