Nex-AGI1.6T (49B active)~967 GB VRAM at Q4_K_M

Nex-N2.5 Max — VRAM, speed & local setup

Written by Jakub Rusinowski · Last updated September 8, 2026

A 1.6-trillion-parameter text-only MoE on a DeepSeek-V4-Pro base — 967 GB at Q4_K_M, which Nex serves on 16x H200 across two nodes. The best BrowseComp score in Nex's table at 92.6, plus 86.1 on Terminal-Bench 2.1 and 65.7 on SWE-Bench Pro. Listed here as a reference point for what frontier open weights now cost to hold, not as a local recommendation. DeepSeek lineage means MLA compressed KV cache, so its context is far cheaper than the parameter count suggests.

Nex-N2.5 Max needs about 967 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.

Logic95

Creative90

Coding95

VRAM and speed by quantization

Quoted against NVIDIA RTX 4090 (24 GB). Weights plus overhead only: this model's architecture is unpublished, so no KV cache is added.

Quant	VRAM	Speed (est.)	Fit
Q2_K 2.63 bpw	526.8 GB	—	Too big
Q3_K_M 3.41 bpw	682.8 GB	—	Too big
Q4_K_M 4.83 bpw	966.8 GB	—	Too big
Q5_K_M 5.67 bpw	1134.8 GB	—	Too big
Q6_K 6.56 bpw	1312.8 GB	—	Too big
Q8_0 8.50 bpw	1700.8 GB	—	Too big
F16 16.00 bpw	3200.8 GB	—	Too big

Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. Nex-N2.5 Max VRAM calculator →

Get Nex-N2.5 Max running

Nex-N2.5 Max is not practical on a single consumer GPU: it needs about 966.8 GB at Q4_K_M, more than any single device in our catalogue holds with headroom. How multi-GPU inference works, or rent a multi-GPU server by the hour:

Deploy in the cloud now RunPod

or compare on Vast.ai

As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.

How to run Nex-N2.5 Max

Install Ollama, then run:

$ollama run nex-n2-5

Weights on Hugging Face: nex-agi/Nex-N2.5-Max ↗

Specifications

Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.

Parameters: 1.6 Trillion (49B active)
Context window: 262,144
Architecture: Mixture-of-Experts (DeepSeek-V4-Pro base, MLA attention, text-only)
Provider: Nex-AGI
Licence: Open-weight (terms unpublished)
Practical deployment: Private cloud / datacenter cluster
Specified at: Q4_K_M
System RAM: 1024 GB
Record updated: 2026-09-08

LicenceCustom Open-WeightCommercial use permitted

Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.

Quality and use cases

Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.

Best forfrontier tasksagentic tasksresearchenterprise

Can I run Nex-N2.5 Max on my GPU?

Other Nex-N2.5 sizes

Nex-N2.5 Max — frequently asked questions

How much VRAM does Nex-N2.5 Max need?

About 967 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.

Does Nex-N2.5 Max run on an RTX 4090 (24 GB)?

No. Nex-N2.5 Max needs about 967 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.

How do I run Nex-N2.5 Max locally?

Install Ollama and run `ollama run nex-n2-5`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does Nex-N2.5 come in?

Nex-N2.5 mini (22 GB), Nex-N2.5 Pro (240 GB), Nex-N2.5 Max (967 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.