whatmodel.sh / Mellum2 12B-A2.5B Thinking

Mellum2 12B-A2.5B Thinking

+ contribute

Mellum2 12B-A2.5B Thinking is an open-weights language model from JetBrains, released May 2026. Its published size is 12.15B parameters (2.5B active per token) and a 131k-token context window. Licence: apache-2.0.

Catalogue identity notes

JetBrains/Mellum2-12B-A2.5B-Thinking @a731155 (created 2026-05-26, licence apache-2.0). JetBrains Mellum2 coding-focused MoE (thinking, May 2026; 64 experts, 8 active, 12B total / 2.5B active per the name). Rows from its own card table with JetBrains' runs of catalogued Qwen3.5 4B/9B, OLMo 3 7B and Ministral 3 14B.

Model information sources

Release date ↗ · Weights licence ↗

HF safetensors metadata 12,149,923,072 parameters; 2.5B active per the maker card.

54.2±1.3#286 of 719 LLMsfrom 18 benchmark results across 2 sources

$ how it compares

The LLMs ranked either side of Mellum2 12B-A2.5B Thinking, out of 828 with a score, then more from JetBrains.

#	model	score	where	RAM	released
338	Bielik 11B v2.6 Instruct SpeakLeash	54±3	local	~8.1GB	2025-06
339	Nova Premier Amazon	54±2	api $5.00/M		2025-04
340	Mellum2 12B-A2.5B Thinking JetBrains	54±1	local	~8.6GB	2026-05
341	Mistral Large 2 (2407) Mistral	54±1	local	~45GB	2024-07
342	Mistral Small 3.1 24B (2503) Mistral	54±1	local	~17GB	2025-03
	more from JetBrains
228	Mellum2.1 12B-A2.5B Thinking JetBrains	60±2	local	~8.3GB	2026-09
410	Mellum2 12B-A2.5B Instruct JetBrains	50±2	local	~8.6GB	2026-05

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #249 of 300 · 24% of its score · 5 tests

Beats DeepSeek R1 0528 Qwen3 8B, Magistral Small 1.0 (2506), Llama 4 Scout (109B-A17B) here, though they rank higher overall. Trails Granite 4.0 H 1B, Granite 4.0 1B, SmolLM3 3B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BFCL V4 (December 2025 archive) lab +1	45.6% ✎	▁▁▁▂▇▆▆▆█▂▄▂ beats 70/111	medium · avg 42%	▇▇▇ 8.0%	at its level · several runs
Terminal-Bench 2.1 lab	0.6% ✎	▃·▁·▁▁▄▃▄█▃▇ beats 0/56	very hard · avg 6.5%	▇▇ 5.7%	typical
SWE-bench Verified lab	2.0% ✎	▁▃▁▁▁▁▂▄█▆▁▁ beats 1/131	hard · avg 17%	▇▇ 5.1%	sharp · noisy
SWE-bench Pro (split unspecified) lab	0.0% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 0/55	very hard · avg 7.5%	▇▇ 4.0%	sharp · noisy
BFCL v3 lab	69.4% ✎	▁▁▁▁▁▃▅█▆▅▄▁ beats 147/170	medium · avg 58%	▇ 1.6%	blunt · at its level

coding▁▁▁▃▆██▇▆▄▃▁ #227 of 458 · 36% of its score · 5 tests

Beats Qwen3.5 4B, Qwen3-VL 32B Instruct, Solar Open 100B (102B-A12B) here, though they rank higher overall. Trails QwQ-32B-Preview, Ministral 3 3B Reasoning 2512, GPT-4 (0613) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench v6 (window unspecified) lab +1	69.9% ✎	▂▃▄▄▅██▅▄▇▃▆ beats 37/65	medium · avg 56%	▇▇▇▇▇▇ 15%	at its level · several runs
HumanEval+ lab	90.9% ✎	▁·▁▁▂▃▄▄█▅▄▃ beats 134/145	easy · avg 80%	▇▇▇▇ 8.8%	sharp
SWE-bench Verified lab	2.0% ✎	▁▃▁▁▁▁▂▄█▆▁▁ beats 1/131	hard · avg 17%	▇▇ 5.1%	sharp · noisy
SWE-bench Pro (split unspecified) lab	0.0% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 0/55	very hard · avg 7.5%	▇▇ 4.0%	sharp · noisy
MBPP+ lab	75.4% ✎	▁··▁▁▃▄█▆▄▁▁ beats 101/120	easy · avg 66%	▇ 3.6%	at its level

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #250 of 393 * · 4.4% of its score · 1 test

Beats Mistral Small 4 (119B-A6.5B), Ministral 3 14B Reasoning 2512, Qwen2.5 Coder 32B Instruct here, though they rank higher overall. Trails Flan-T5 XXL, Granite 4.0 H 1B, Granite 4.0 1B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
IFEval (aggregation unspecified) lab +1	76.5% ✎	▁▁▃▁▂▄▄▆█▃▂▁ beats 51/141	easy · avg 81%	▇▇ 4.4%	several runs

other tests3 not in a task · 43% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond lab +1	57.6% ✎	▁▁▁▁▂▄▅▅███▄ beats 106/310	hard · avg 44%	▇▇▇▇▇▇▇▇▇▇ 24%	sharp · at its level · several runs
MMLU-Redux lab +1	86.2% ✎	▁▁▃▁▂▃▄▅▇▅█▄ beats 24/63	easy · avg 79%	▇▇▇▇▇▇ 14%	several runs
GSM-Plus lab +1	87.0% ✎	▁··▁▁▁·▃▁▄█▁ beats 10/16	easy · avg 83%	▇▇ 5.0%	blunt · several runs

$ where to run it

llama.cpp · recommended

version

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run Mellum2 12B-A2.5B Thinking · needs ~8.2 GB · from huggingface.co/bartowski
chat in the terminal
llama cli -hf bartowski/Mellum2-12B-A2.5B-Thinking-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf bartowski/Mellum2-12B-A2.5B-Thinking-GGUF:Q4_K_M

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Mellum2 12B-A2.5B Thinking · needs ~8.2 GB · from huggingface.co/bartowski
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/bartowski/Mellum2-12B-A2.5B-Thinking-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Mellum2 12B-A2.5B Thinking · needs ~8.2 GB · from huggingface.co/bartowski
chat in the terminal
ollama run hf.co/bartowski/Mellum2-12B-A2.5B-Thinking-GGUF:Q4_K_M

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Mellum2 12B-A2.5B Thinking · needs ~6.8 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/Mellum2-12B-A2.5B-Thinking-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/Mellum2-12B-A2.5B-Thinking-4bit

Other ways to run it

Run it as published · needs ~24.3 GB, NVIDIA only · from huggingface.co/JetBrains (official)
vllm serve JetBrains/Mellum2-12B-A2.5B-Thinking

Official weights and documentation: huggingface.co/JetBrains ↗

$ how the score was worked out · JetBrains/Mellum2.1-12B-A2.5B… 73%, JetBrains/Mellum2-12B-A2.5B-T… 27%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The 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's added up by who ran the tests. How it works

source	results	share of the score	licence
JetBrains/Mellum2.1-12B-A2.5B-Thinking model card · lab-reported	11	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 73%	Apache-2.0
JetBrains/Mellum2-12B-A2.5B-Thinking model card · lab-reported	7	▇▇▇▇▇▇▇▇ 27%	Apache-2.0
prior · pull toward the middle		▇ 1%
weighted average		= 54.2

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-10-11 02:20 UTC · sources and licences · json