readygpu

Mellum2.1 12B A2.5B Thinking

12.1B total, 2.5B active · 128k context · Apache 2.0 · released 2026-09-20

chatcodingagenticreasoning

In short

Mellum2.1 12B A2.5B Thinking is an open model from JetBrains with 12.1B parameters (2.5B active per token). At Q6_K, Mellum2.1 12B A2.5B Thinking needs about 12 GB of memory at 8k context and 13 GB at 64k, counting the weights, the KV cache and runtime buffers. The cheapest hardware in Germany that runs Mellum2.1 12B A2.5B Thinking fully in memory at 64k context is the NVIDIA Tesla P40, 369 € at eBay, at about 35 tok/s (estimated).

Updated . Speeds are estimates; how they are worked out.

What to run it on in Germany

Fully in memory at 64k context, priced at retailers today.

All hardware that runs it

Everything we price, best value first. Two cards split the model by layers.

HardwareQuantFitsDecodePricetok/s per 1,000
NVIDIA Tesla P40Q8_0fits35 tok/s369 €95.7
AMD Radeon RX 9060 XT 16GBQ8_0fits29 tok/s550 €52.7
Intel Arc A770 16GBQ8_0fits42 tok/s646 €65.5
NVIDIA GeForce RTX 4060 Ti 16GBQ8_0fits29 tok/s650 €45.2
NVIDIA GeForce RTX 5060 Ti 16GBQ8_0fits46 tok/s739 €61.8
AMD Radeon RX 9070 XTQ8_0fits58 tok/s778 €74.5
AMD Radeon RX 7600 XTQ8_0fits26 tok/s813 €32.1
AMD Radeon RX 7900 XTQ8_0fits73 tok/s855 €84.8
AMD Radeon RX 7800 XTQ8_0fits57 tok/s1.002 €56.4
Intel Arc Pro B60Q8_0fits34 tok/s1.074 €32.0
NVIDIA GeForce RTX 4080 SuperQ8_0fits75 tok/s1.100 €68.4
NVIDIA GeForce RTX 5070 TiQ8_0fits92 tok/s1.149 €79.6
NVIDIA GeForce RTX 4070 Ti SuperQ8_0fits69 tok/s1.247 €55.0
Mac mini M4 32GBQ8_0fits13 tok/s1.400 €9.1
NVIDIA GeForce RTX 5080Q8_0fits98 tok/s1.500 €65.3
AMD Radeon RX 7900 XTXQ8_0fits87 tok/s1.555 €55.9
NVIDIA GeForce RTX 3090Q8_0fits96 tok/s1.850 €51.7
AMD Radeon AI PRO R9700Q8_0fits58 tok/s1.867 €31.1
NVIDIA GeForce RTX 3090 TiQ8_0fits103 tok/s2.500 €41.2
Mac mini M4 Pro 48GBQ8_0fits29 tok/s2.699 €10.8
Mac mini M4 Pro 64GBQ8_0fits29 tok/s2.799 €10.4
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBQ8_0fits23 tok/s3.700 €6.3
NVIDIA GeForce RTX 4090Q8_0fits103 tok/s3.955 €26.0
Mac Studio M4 Max 64GBQ8_0fits58 tok/s4.183 €14.0
Mac Studio M4 Max 128GBQ8_0fits58 tok/s5.800 €10.1
Mac Studio M3 Ultra 96GBQ8_0fits88 tok/s5.999 €14.6
NVIDIA GeForce RTX 5090Q8_0fits183 tok/s6.028 €30.4
NVIDIA DGX Spark 128GBQ8_0fits28 tok/s8.313 €3.4
ASUS Ascent GX10 128GBQ8_0fits28 tok/s8.623 €3.2
Mac Studio M3 Ultra 256GBQ8_0fits88 tok/s15.555 €5.6
Mac Studio M3 Ultra 512GBQ8_0fits88 tok/s18.995 €4.6
NVIDIA RTX PRO 6000 BlackwellQ8_0fits183 tok/s19.076 €9.6
Framework Desktop Ryzen AI Max+ 395 128GBQ8_0fits23 tok/s––
Framework Desktop Ryzen AI Max+ 395 64GBQ8_0fits23 tok/s––
NVIDIA GeForce RTX 3060 12GBQ6_Koffload30 tok/s309 €96.8
Intel Arc B580Q6_Koffload26 tok/s360 €72.6
NVIDIA GeForce RTX 3080 10GBQ6_Koffload27 tok/s500 €53.2
NVIDIA GeForce RTX 4070 SuperQ6_Koffload37 tok/s707 €52.2
NVIDIA GeForce RTX 5070Q6_Koffload43 tok/s798 €54.1

1 other card or machine we price cannot hold it at this context.

Memory needed

Weights, KV cache and runtime buffers at each context length (KV cache at fp16).

QuantFile@8k@32k@64k@128k
Q6_Kgguf10.7 GB11.6 GB12.0 GB12.7 GB13.9 GB
Q8_0gguf12.0 GB12.9 GB13.4 GB14.0 GB15.3 GB

Run it

Copy a command for the runtime you use.

  • Q6_K
    • Ollamaollama run hf.co/bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q6_K
    • LM Studiolms get bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF@q6_k
    • llama.cppllama-server -hf bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q6_K
  • Q8_0
    • Ollamaollama run hf.co/bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q8_0
    • LM Studiolms get bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF@q8_0
    • llama.cppllama-server -hf bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q8_0

Notes

Thinking model; serve with vLLM using the qwen3 reasoning parser (and hermes tool-call parser for tools). GGUF builds and the MTP head are not yet released.

Questions

How much VRAM does Mellum2.1 12B A2.5B Thinking need?
At Q6_K, Mellum2.1 12B A2.5B Thinking needs about 12 GB of memory at 8k context and 13 GB at 64k, counting the weights, the KV cache and runtime buffers.
What is the best GPU for running Mellum2.1 12B A2.5B Thinking locally?
The cheapest hardware in Germany that runs Mellum2.1 12B A2.5B Thinking fully in memory at 64k context is the NVIDIA Tesla P40, 369 € at eBay, at about 35 tok/s (estimated). The fastest we price is the NVIDIA GeForce RTX 5090, at about 183 tok/s.
Can I run Mellum2.1 12B A2.5B Thinking on a 16 GB GPU?
Yes: at Q8_0 it fits fully in 16 GB at 64k context. On the NVIDIA GeForce RTX 5080 it runs at about 98 tok/s (estimated).
Can I run Mellum2.1 12B A2.5B Thinking on a 24 GB GPU?
Yes: at Q8_0 it fits fully in 24 GB at 64k context. On the NVIDIA GeForce RTX 3090 Ti it runs at about 103 tok/s (estimated).
How do I run Mellum2.1 12B A2.5B Thinking locally?
With Ollama: ollama run hf.co/bartowski/Mellum2.1-12B-A2.5B-Thinking-GGUF:Q6_K (the Q6_K build). LM Studio and llama.cpp commands are on this page.
What licence is Mellum2.1 12B A2.5B Thinking released under?
Apache 2.0, which allows commercial use.