[whatmodel.sh](../../README.md) / Mellum2.1 12B-A2.5B Thinking # Mellum2.1 12B-A2.5B Thinking JetBrains · released 2026-09 · 12.15B params (2.5B active) · apache-2.0 **score 60.0** ±1.4 · #202 of 732 · from 11 benchmark results across 1 source · † no independently run results verified Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at ● marks Mellum2.1 12B-A2.5B Thinking among every model on that task or test, worst to best. 
task            vs the others          rank  counts  tests
agent tasks     ▁▁▃▆█●▅▅▄▄▂▁    #156 of 313     34%      4
coding          ▁▁▁▃▅▇█●▅▄▃▁    #171 of 469     40%      5
chat & writing  ▁▁▁▃▆▄▆█●▇▇▃   #169 of 402*    3.3%      1
agent tasks · #156 of 313 · 34% of its score Beats [o1 (2024-12-17)](o1-2024-12-17.md), [o4-mini](o4-mini.md), [Qwen3 235B-A22B Thinking 2507](qwen3-235b-a22b-thinking-2507.md) here, though they rank higher overall. Trails [SmolLM3 3B](smollm3-3b.md), [Jamba 1.5 Mini](jamba-1-5-mini.md), [Jamba 1.5 Large](jamba-1-5-large.md) here, though they rank lower overall. 
test                                result  vs the others in its run   difficulty            counts  why
Terminal-Bench 2.1 (maker)           17.4%  ▁·▁·▁●▄▃▆█▅█ beats 4/63    very hard · avg 6.2%     10%  sharp
SWE-bench Verified (maker)           47.0%  ▁▁▁▁▁▁▂●█▆▁▁ beats 37/138  hard · avg 16%          8.6%  sharp · at its level · noisy
SWE-bench Pro (split unspecified)…   28.0%  ▂·▁▁▁▁●▃█▇▂▁ beats 14/58   very hard · avg 7.2%    8.6%  sharp · noisy
BFCL V4 (December 2025 archive) (…   62.3%  ▁▁▁▂▆▆▅▆█●▄▃ beats 96/114  medium · avg 42%        6.3%  at its level
coding · #171 of 469 · 40% of its score Beats [Qwen3-Coder-Next (80B-A3B)](qwen3-coder-next.md), [Gemini 2.5 Pro](gemini-2-5-pro.md), [Grok 4.1 Fast (Reasoning)](grok-4-1-fast-reasoning.md) here, though they rank higher overall. Trails [QwQ-32B-Preview](qwq-32b-preview.md), [Ministral 3 8B Reasoning 2512](ministral-3-8b-reasoning.md), [Phi-4-reasoning 14B](phi-4-reasoning.md) here, though they rank lower overall. 
test                                result  vs the others in its run    difficulty            counts  why
LiveCodeBench v6 (window unspecif…   82.0%  ▂▃▄▄▅██▆●▇▄▆ beats 48/67    medium · avg 56%         11%  typical
SWE-bench Verified (maker)           47.0%  ▁▁▁▁▁▁▂●█▆▁▁ beats 37/138   hard · avg 16%          8.6%  sharp · at its level · noisy
SWE-bench Pro (split unspecified)…   28.0%  ▂·▁▁▁▁●▃█▇▂▁ beats 14/58    very hard · avg 7.2%    8.6%  sharp · noisy
HumanEval+ (maker)                   91.5%  ▁·▁▁▂▃▄▄█▅▄● beats 136/145  easy · avg 80%          7.9%  sharp · maxed out
MBPP+ (maker)                        79.4%  ▁··▁▁▃▄█▆●▁▁ beats 108/120  medium · avg 65%        4.2%  typical
chat & writing · #169 of 402 · 3.3% of its score Beats [MiniMax-M2](minimax-m2.md), [MiniCPM5 2B](minicpm5-2b.md), [Granite 4.2 30B](granite-4-2-30b.md) here, though they rank higher overall. Trails [LFM2.5 1.2B Instruct](lfm2-5-1-2b-instruct.md), [LFM2.5 1.2B Thinking](lfm2-5-1-2b-thinking.md), [LFM2.5 8B-A1B](lfm2-5-8b-a1b.md) here, though they rank lower overall. 
test                                result  vs the others in its run    difficulty      counts  why
IFEval (aggregation unspecified) …   90.6%  ▁▂▄▂▂▄▆█●▅▂▁ beats 119/145  easy · avg 81%    3.3%  typical
other tests (3) 
test                  result  vs the others in its run    difficulty      counts  why
GPQA Diamond (maker)   64.6%  ▁▁▁▁▂▄▅●███▄ beats 131/322  hard · avg 43%     27%  sharp · at its level
MMLU-Redux (maker)     87.8%  ▁▁▃▁▂▃▄●▇▅█▄ beats 27/63    easy · avg 79%    8.7%  typical
GSM-Plus (maker)       88.3%  ▁··▁▁▁·▃▁▄●▁ beats 14/16    easy · avg 83%    2.8%  blunt
## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its *counts* column above; here it is added up by who ran the tests. [Method](../method.md) 
source                                    results  share of the score
JetBrains/Mellum2.1-12B-A2.5B-Thinking …       11   99% ▇▇▇▇▇▇▇▇▇▇▇▇
prior: pull toward the middle                        1% ▇
weighted average                                   = 60.0
Licences: JetBrains/Mellum2.1-12B-A2.5B-Thinking model card ([Apache-2.0](https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking/blob/92ddae9fc7665e9f801d141d2e5a6b2caf2460c4/README.md)) ## How to run it ### llama.cpp (recommended) 1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md) ```sh # macOS / Linux curl -LsSf https://llama.app/install.sh | sh # Windows irm https://llama.app/install.ps1 | iex ``` 2. Run Mellum2.1 12B-A2.5B Thinking (Q4_K_M, needs ~8.1 GB, from [huggingface.co/JetBrains](https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF/tree/20b439426f1e997ba575eaccaf4856ac0f76f5d5)) ```sh llama cli -hf JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q4_K_M ``` 3. Or start a local server and open the address it prints (http://localhost:9931) ```sh llama serve -hf JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q4_K_M ``` Other versions: Q8_0 (~12.9 GB) `llama cli -hf JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q8_0` ### LM Studio 1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download) 2. Run Mellum2.1 12B-A2.5B Thinking (Q4_K_M, needs ~8.1 GB, from [huggingface.co/JetBrains](https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF/tree/20b439426f1e997ba575eaccaf4856ac0f76f5d5)) ```sh lms get https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF@Q4_K_M -y ``` 3. Then run lms ls to see its name, and lms chat followed by that name Other versions: Q8_0 (~12.9 GB) `lms get https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF@Q8_0 -y` · mxfp4 (~6.5 GB, Mac only) `lms get https://huggingface.co/randmaru/Mellum2.1-12B-A2.5B-Thinking-mlx-mxfp4 -y` ### Ollama 1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download) ```sh # Linux curl -fsSL https://ollama.com/install.sh | sh ``` 2. Run Mellum2.1 12B-A2.5B Thinking (Q4_K_M, needs ~8.1 GB, from [huggingface.co/JetBrains](https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF/tree/20b439426f1e997ba575eaccaf4856ac0f76f5d5)) ```sh ollama run hf.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q4_K_M ``` Other versions: Q8_0 (~12.9 GB) `ollama run hf.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q8_0` ### MLX (Mac) 1. Install MLX (Mac). Apple silicon Macs only, text only. ```sh # with uv uv tool install mlx-lm # or pip install mlx-lm ``` 2. Run Mellum2.1 12B-A2.5B Thinking (mxfp4, needs ~6.5 GB, Mac only, from [huggingface.co/randmaru](https://huggingface.co/randmaru/Mellum2.1-12B-A2.5B-Thinking-mlx-mxfp4/tree/96b37c8dd154d230ddb5fcc601405b496e3bafe8)) ```sh mlx_lm.chat --model randmaru/Mellum2.1-12B-A2.5B-Thinking-mlx-mxfp4 ``` 3. Or start an OpenAI-compatible server at http://localhost:8080/v1 ```sh mlx_lm.server --model randmaru/Mellum2.1-12B-A2.5B-Thinking-mlx-mxfp4 ``` ### Other ways to run it 1. Run it as published (BF16, needs ~24.3 GB, NVIDIA only, from [huggingface.co/JetBrains](https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking/tree/92ddae9fc7665e9f801d141d2e5a6b2caf2460c4)) ```sh vllm serve JetBrains/Mellum2.1-12B-A2.5B-Thinking ``` Model repository and documentation: [huggingface.co/JetBrains](https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking) [interactive version](https://whatmodel.sh/model/mellum2-1-12b-a2-5b-thinking) · [JSON](https://whatmodel.sh/api?format=json&all=1)