DeepSeekDeepSeek V2DeepseekV2ForCausalLM

DeepSeek V2.5 — Hardware Requirements & GPU Compatibility

ChatFunctions

DeepSeek-V2.5 is DeepSeek's merged general-purpose and coding chat model, combining DeepSeek-V2-Chat and DeepSeek-Coder-V2-Instruct into a single mixture-of-experts checkpoint with about 235.7 billion total parameters and roughly 21.4 billion active per token. DeepSeek reports it better aligns with human preferences and improves writing and instruction-following over its two predecessors, with gains on AlpacaEval, ArenaHard, and coding benchmarks like HumanEval and LiveCodeBench. It is a chat, not base, model. Given its total parameter count, DeepSeek states that running it in BF16 requires eight 80GB-class GPUs, so it needs a substantial multi-GPU server, not a single consumer machine, even once quantized. Context length is 163,840 tokens. Model weights are released under DeepSeek's custom model license, which permits commercial use, while the surrounding code is MIT-licensed. It was published in September 2024, unifying the separate V2-Chat and Coder-V2-Instruct lines into one model.

235.7B params · MoE21B active164K contextOtherSep 20244.5K downloads750 likes4.9K quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

DeepSeek V2.5 runs on a 192 GB+ Mac or a datacenter GPU

Best local pick → Mac Pro M2 Ultra (192 GB) · Q4_K_M · 142.9 GB · ~28 tok/s

Macs from 96 GB (IQ2_M · ~40 tok/s)

Other

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run DeepSeek V2.5

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Recommended

Ollama

one command · needs Ollama installed

$ollama run hf.co/bartowski/DeepSeek-V2.5-GGUF:Q4_K_M

Q4_K_M runs on one GH200 Grace Hopper Superchip 144 GB · 16K context

Download: DeepSeek-V2.5-Q4_K_M/DeepSeek-V2.5-Q4_K_M-00001-of-00004.gguf142.45 GB + 3 partsbartowski

install, OS, other quants

Ways to runOther ways

each opens step-by-step

Run in cloud

Your own private copy · Q4_K_M · 96 GB GPU

The easiest way to try this model
Running in a few minutes, easy setup
Chat right here or connect any app

$5+ bonus for new users

How Much VRAM Does DeepSeek V2.5 Need?

Select a quantization level; the listed hardware adheres to it.

Quantization	VRAM	+32K	+128K	File	Quality	Runs on
	67.7 GB	67.4 GB	74.2 GB	61.5 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Runs on RTX PRO 5000 Blackwell 72 GB
	73.5 GB	73.3 GB	80.1 GB	68.7 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Runs on H100 SXM 80 GB
	82.4 GB	82.1 GB	88.9 GB	76.9 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
	94.2 GB	93.9 GB	100.7 GB	90.8 GB	3-bit · i-quant — Importance-weighted 3-bit	Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
	103.0 GB	102.8 GB	109.6 GB	85.9 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Runs on Instinct MI250X 128 GB
	106.0 GB	105.7 GB	112.5 GB	101.7 GB	3-bit small — 3-bit small quantization	Runs on Instinct MI250X 128 GB
	108.9 GB	108.7 GB	115.4 GB	103.4 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Runs on Instinct MI250X 128 GB
	117.7 GB	117.5 GB	124.3 GB	112.7 GB	3-bit medium — 3-bit medium quantization	Runs on Instinct MI250X 128 GB
	120.7 GB	120.4 GB	127.2 GB	133.4 GB	4-bit · legacy — 4-bit legacy quantization	Runs on Instinct MI250X 128 GB
	123.6 GB	123.4 GB	130.2 GB	122.4 GB	3-bit large — 3-bit large quantization	Runs on Instinct MI250X 128 GB
	129.5 GB	129.3 GB	136.1 GB	125.6 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Runs on H200 NVL 141 GB
	135.4 GB	135.2 GB	142 GB	133.9 GB	4-bit small — 4-bit small quantization	Runs on H200 NVL 141 GB
	144.3 GB	144 GB	150.8 GB	142.5 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Runs on GH200 Grace Hopper Superchip 144 GB
	150.2 GB	149.9 GB	156.7 GB	162.3 GB	5-bit · legacy — 5-bit legacy quantization	Runs on B200 192 GB
	164.9 GB	164.6 GB	171.4 GB	162.3 GB	5-bit small — 5-bit small quantization	Runs on B200 192 GB
	170.8 GB	170.5 GB	177.3 GB	167.2 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Runs on B200 192 GB
	197.3 GB	197.1 GB	203.8 GB	193.5 GB	6-bit, very good — 6-bit quantization, very good quality	Runs on Instinct MI325X 256 GB
	238.6 GB	238.3 GB	245.1 GB	250.6 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Runs on Instinct MI325X 256 GB
	474.3 GB	474.1 GB	480.8 GB	~471.5 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Runs on 7× H100 SXM 80 GB

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

142.9GBat 16K context

Weights: 141.4 GB
KV cache: 1.1 GB
Overhead: 0.3 GB

Server-class: ~142.9 GB at 16K context · one GH200 Grace Hopper Superchip 144 GB

Or .

Which Hardware Can Run DeepSeek V2.5?

Q4_K_M · 144.3 GB

DeepSeek V2.5 (Q4_K_M) requires 144.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 188+ GB is recommended. Using the full 164K context window can add up to 8.8 GB, bringing total usage to 153.1 GB. No consumer GPU has enough memory. 4 devices with unified memory can run DeepSeek V2.5, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (192 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 66.3 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →

How good is DeepSeek V2.5?

llmrun Score blends public benchmark boards into one number; 100 = the average open model.

Not enough data yet for an llmrun Score.

A Score needs results on enough independent boards. How the Composite Score is built

Benchmarks

LMArena Text
LMArena · Oct 2026Arena
1307.2
#102/188 open
Arena
MMLU-Pro
TIGER-LabIndependent
65.8%
#46/146 open
Independent
Aider Polyglot
Epoch AI · Sep 2024Independent
17.8%
#13/18 open
Independent

All 3 DeepSeek V2.5 benchmarks

Alternatives to DeepSeek V2.5

DeepSeek V2 family · 5 models →

Similar models

5

Model	Score	VRAM	tok/s	Why it is similar
MiniMax M2.7MiniMax	107.4	139.6 GB	—needs > 24 GB	same size
Command A Plus 05 2026 BF16Cohere	—	132.1 GB	—needs > 24 GB	same size
Qwen3 VL 235B A22B InstructAlibaba	—	143.3 GB	—needs > 24 GB	same size
Laguna M.1poolside	—	138.1 GB	—needs > 24 GB	same size
Llama 3 1 Nemotron Ultra 253B V1NVIDIA	—	564.2 GB	—needs > 24 GB	same size

Specifications

Publisher: DeepSeek
Family: DeepSeek V2
Parameters: 235.7B
Active parameters: 21.4B per token
Architecture: DeepseekV2ForCausalLM
Context Length: 163,840 tokens
Vocabulary Size: 102,400
Release Date: 2024-09-05
License: Other

Resources

Frequently Asked Questions

How much VRAM does DeepSeek V2.5 need?

DeepSeek V2.5 requires 144.3 GB of VRAM at Q4_K_M, or 474.3 GB at BF16. Full 164K context adds up to 8.8 GB (153.1 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 235.7B × 4.8 bits ÷ 8 = 141.4 GB

KV Cache + Overhead ≈ 2.9 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 11.7 GB (at full 164K context)

VRAM usage by quantization

Q4_K_M

144.3 GB

Q4_K_M + full context

153.1 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run DeepSeek V2.5?

No — DeepSeek V2.5 requires at least 67.7 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for DeepSeek V2.5?

For DeepSeek V2.5, Q4_K_M (144.3 GB) offers the best balance of quality and VRAM usage. Q5_0 (150.2 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 67.7 GB.

VRAM requirement by quantization

IQ2_XXS

67.7 GB~53%

Q3_K_S

106.0 GB~77%

Q3_K_L

123.6 GB~86%

Q4_K_M ★

144.3 GB~89%

Q5_K_S

164.9 GB~92%

BF16

474.3 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run DeepSeek V2.5 on a Mac?

Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run DeepSeek V2.5, the smallest at IQ2_M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 144.3 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (192 GB).

Can I run DeepSeek V2.5 locally?

Yes — DeepSeek V2.5 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is DeepSeek V2.5?

At Q4_K_M, DeepSeek V2.5 can reach ~68 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)

Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.

Example: NVIDIA B200 → 13.1 GB active ÷ (8000 × 0.65) = 2.52 ms, plus 60 layers × 0.055 ms = 3.30 ms, so 1000 ÷ 5.82 ms = ~172 tok/s

Estimated speed at Q4_K_M (144.3 GB)

NVIDIA B200

~172 tok/s

NVIDIA B300

~172 tok/s

AMD Instinct MI350X

~68 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of DeepSeek V2.5?

At Q4_K_M, the download is about 141.44 GB. The full-precision BF16 version is 471.48 GB. The smallest option (IQ2_XXS) is 64.83 GB.

Which GPUs can run DeepSeek V2.5?

No single consumer GPU has enough VRAM to run DeepSeek V2.5 at Q4_K_M (144.3 GB). Multi-GPU or professional hardware is required.

Which devices can run DeepSeek V2.5?

4 devices with unified memory can run DeepSeek V2.5 at Q4_K_M (144.3 GB), including Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M2 Ultra (192 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.

Run DeepSeek V2.5

142.9 GB VRAM at 16K context · runs on one GH200 Grace Hopper Superchip 144 GB

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/bartowski/DeepSeek-V2.5-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/bartowski/DeepSeek-V2.5-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/bartowski/DeepSeek-V2.5-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf bartowski/DeepSeek-V2.5-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/bartowski/DeepSeek-V2.5-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf bartowski/DeepSeek-V2.5-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1. Splits the model over 8 GPUs of one server (tensor parallel).
$vllm serve deepseek-ai/DeepSeek-V2.5 --max-model-len 16384 --tensor-parallel-size 8
Runs the original weights, not the Q4_K_M GGUF; needs ~519 GB of GPU memory.

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$bartowski/DeepSeek-V2.5-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
B200 (192 GB)CheapestCheapest49 GB free	~172 tok/s	$9.66	Deploy on RunPodfrom $5.98/hr (opens in new tab)	192 GB · 49 GB free · ~172 tok/s · $9.66 / 1M tokens
B300 (288 GB)145 GB free	~172 tok/s	$11.21	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 145 GB free · ~172 tok/s · $11.21 / 1M tokens
B200 (192 GB)49 GB free	~172 tok/s	$12.12	Deploy on Vast.aifrom $7.50/hr (opens in new tab)	192 GB · 49 GB free · ~172 tok/s · $12.12 / 1M tokens

Compare GPUs →

Your own private copy of DeepSeek V2.5 · Q4_K_M · OpenAI-compatible API

The pod’s first start downloads ~141 GB, billed while it loads.

Use it

RunPod → your pod → Connect → HTTP 8080, copy the link. The key is the one you made up in step 2.

Endpoint URL

API key

Chats and keys stay only in this browser.

https://<POD_ID>-8080.proxy.runpod.net/v1

RunPod → your pod → Connect → HTTP 8080 gives this address. The API key is the one from step 2.

from openai import OpenAI
client = OpenAI(base_url="https://<POD_ID>-8080.proxy.runpod.net/v1", api_key="YOUR_KEY")
r = client.chat.completions.create(
    model="any",
    messages=[{"role": "user", "content": "Hello!"}])
print(r.choices[0].message.content)

Works with any OpenAI client.

https://<POD_ID>-8080.proxy.runpod.net/v1

RunPod → your pod → Connect → HTTP 8080 gives this address. The API key is the one from step 2.

Free on Windows, macOS, Linux, iOS, Android and the web.

Open the web app ↗ (opens in a new tab)Download ↗ (opens in a new tab)

1. Settings → Model Provider → Add → Add Custom Provider
2. API Mode OpenAI API Compatible; paste the API key and the base URL as API Host
3. Model → Fetch, pick the model
4. New chat → pick the provider → start chatting

Free and open source for Windows, macOS and Linux.

Download Jan ↗ (opens in a new tab)

1.Settings → next to Model Providers press +
2.API format OpenAI-compatible; Base URL is the base URL above (with /v1); API key is your key; Create
3.If no model appears, click + in Models and type any
4.New chat → pick the model → start chatting