Is Mellum2-12B-A2.5B-Instruct-MXFP4_MOE good for coding?

In our benchmarks, Mellum2-12B-A2.5B-Instruct-MXFP4_MOE scores 59.1/100 for coding. It runs at about 86.6 tok/s, so if you want more speed, Ornith-1.5-35B-A3B.APEX-I-NanoPlus is faster (~110.5 tok/s) and still scores well for coding (81.1/100).

Is Mellum2-12B-A2.5B-Instruct-MXFP4_MOE good for agentic (tool-using) tasks?

In our benchmarks, Mellum2-12B-A2.5B-Instruct-MXFP4_MOE scores 72.6/100 for agentic workflows. It runs at about 86.6 tok/s, so if you want more speed, qwen3.8-flash-next-iq3_s is faster (~104.5 tok/s) and still scores well for agentic workflows (90.0/100).

How fast is Mellum2-12B-A2.5B-Instruct-MXFP4_MOE for local inference?

Across 10 community benchmark runs, Mellum2-12B-A2.5B-Instruct-MXFP4_MOE reaches up to 88.5 tok/s and averages 86.6 tok/s, with the fastest results on NVIDIA GeForce RTX 4060 Laptop GPU.

How much memory does Mellum2-12B-A2.5B-Instruct-MXFP4_MOE need?

The leanest observed configuration used about 7.4 GB of memory.

Which tools have been used to run Mellum2-12B-A2.5B-Instruct-MXFP4_MOE?

Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.

Model benchmarks

Mellum2-12B-A2.5B-Instruct-MXFP4_MOE local LLM performance

As of October 2026, Mellum2-12B-A2.5B-Instruct-MXFP4_MOE runs at up to 88.5 tok/s for local inference (best of 10 community benchmark runs across 1 GPU).

llama.cpp

ShareReddit X

Model size

12B

Peak speed

88.5 tok/s

Average speed

86.6 tok/s

Avg PP

796.2 tok/s

Min memory

7.4 GB

Max context

32,768 tokens

Avg output / run

5,831 tokens

Avg runtime / run

1m 12s

Avg quality

66.2

Benchmark runs

GPUs tested

Quality by task

Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 10 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.

Task	P5 (low)	Avg	P95 (high)
Overall	63.0	66.2	69.8
Agent Workflow	58.6	72.6	86.2
Code Generation	52.7	59.1	65.8
Role Play & Narrative	61.2	72.9	81.8
Research & Analysis	47.9	60.2	68.3

Performance by hardware and tool

Every hardware/tool/quantization combination Mellum2-12B-A2.5B-Instruct-MXFP4_MOE has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.

Hardware	Tool	Quant	Best tok/s	Avg tok/s	Memory	Context	Quality	Runs
NVIDIA GeForce RTX 4060 Laptop GPU	llama.cpp	—	88.5 tok/s	86.6 tok/s	7.4 GB	32,768 tokens	66.2	10

Benchmark runs

All 10 Mellum2-12B-A2.5B-Instruct-MXFP4_MOE runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is Mellum2-12B-A2.5B-Instruct-MXFP4_MOE good for coding?: In our benchmarks, Mellum2-12B-A2.5B-Instruct-MXFP4_MOE scores 59.1/100 for coding. It runs at about 86.6 tok/s, so if you want more speed, Ornith-1.5-35B-A3B.APEX-I-NanoPlus is faster (~110.5 tok/s) and still scores well for coding (81.1/100).
Is Mellum2-12B-A2.5B-Instruct-MXFP4_MOE good for agentic (tool-using) tasks?: In our benchmarks, Mellum2-12B-A2.5B-Instruct-MXFP4_MOE scores 72.6/100 for agentic workflows. It runs at about 86.6 tok/s, so if you want more speed, qwen3.8-flash-next-iq3_s is faster (~104.5 tok/s) and still scores well for agentic workflows (90.0/100).
How fast is Mellum2-12B-A2.5B-Instruct-MXFP4_MOE for local inference?: Across 10 community benchmark runs, Mellum2-12B-A2.5B-Instruct-MXFP4_MOE reaches up to 88.5 tok/s and averages 86.6 tok/s, with the fastest results on NVIDIA GeForce RTX 4060 Laptop GPU.
How much memory does Mellum2-12B-A2.5B-Instruct-MXFP4_MOE need?: The leanest observed configuration used about 7.4 GB of memory.
Which tools have been used to run Mellum2-12B-A2.5B-Instruct-MXFP4_MOE?: Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.