DeepSeekDeepSeek V4DeepseekV4ForCausalLM

DeepSeek V4 Flash 0731 — Hardware Requirements & GPU Compatibility

Chat

DeepSeek V4 Flash 0731 is a large-scale model from DeepSeek's V4 line, weighing in at roughly 304 billion parameters. At this scale it needs multi-GPU or server-class hardware to run; most users will reach it through a hosted endpoint rather than loading it locally on consumer equipment. It is tuned for general chat and instruction-following use. Its standout feature is a 1M token context window, among the largest available in an open-weight model, enabling it to process entire codebases, books, or long conversation histories in a single pass. Released under the permissive MIT license on July 31, 2026, DeepSeek V4 Flash 0731 is aimed at deployments where raw context length and model capacity matter more than local accessibility.

304.2B params · MoE27B active1049K contextMITJul 20264.4M downloads4.1K likes156.1K 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 V4 Flash 0731 runs on a 192 GB+ Mac or a datacenter GPU

Best local pick → Mac Pro M2 Ultra (192 GB) · Q4_K_M · 183 GB · ~26 tok/s

Macs from 96 GB (IQ2_XS · ~44 tok/s)

Score 127.6#7 of 100 open#2 of 78 that fit ≤192 GBMIT

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

llmrun Composite Score →

127.6

90% range 125.8–129.4
#7 of 100 open models
#2 of 78 that fit ≤192 GB

Benchmarks · tier chart ↓

Among 100 open modelsHow the Composite Score is built

Coding120.890% range 114–127
AgentsNot enough data yet
Math128.190% range 126–131
Science124.390% range 112–136
Reasoning128.790% range 126–132

Run DeepSeek V4 Flash 0731

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on B300 (288 GB) (105 GB headroom) · Q4_K_M

Generation speed: ~181 tok/s; generation speed
Cost per 1M output tokens: $10.66; per 1M output tokens

Deployon RunPodBest optionfrom $6.94/hr

Compare GPUs →

Run DeepSeek V4 Flash 0731

183.0 GB VRAM at 16K context · runs on one B200 192 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/unsloth/DeepSeek-V4-Flash-0731-GGUF
unsloth/DeepSeek-V4-Flash-0731-GGUF does not list Q4_K_M; Ollama picks a quant itself (Q4_K_M if present). Check the repo's Files tab.
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/unsloth/DeepSeek-V4-Flash-0731-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/unsloth/DeepSeek-V4-Flash-0731-GGUF
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 unsloth/DeepSeek-V4-Flash-0731-GGUF -c 16384 -ngl 99 --port 8080
unsloth/DeepSeek-V4-Flash-0731-GGUF does not list Q4_K_M; llama.cpp falls back to Q4_K_M or the first file.

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/unsloth/DeepSeek-V4-Flash-0731-GGUF
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 unsloth/DeepSeek-V4-Flash-0731-GGUF -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-V4-Flash-0731 --max-model-len 16384 --tensor-parallel-size 8
Runs the original weights, not the Q4_K_M GGUF; needs ~669 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model inferencerlabs/DeepSeek-V4-Flash-0731-MLX-Q3i
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model inferencerlabs/DeepSeek-V4-Flash-0731-MLX-Q3i

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.
$unsloth/DeepSeek-V4-Flash-0731-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
B300 (288 GB)CheapestCheapest105 GB free	~181 tok/s	$10.66	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 105 GB free · ~181 tok/s · $10.66 / 1M tokens

Compare GPUs →

How Much VRAM Does DeepSeek V4 Flash 0731 Need?

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

VRAM	+32K	+128K	File	Quality	Runs on
84.0 GB	84.3 GB	85.4 GB	90.9 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
91.6 GB	91.9 GB	93 GB	88.1 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
95.4 GB	95.7 GB	96.8 GB	93.4 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB
103.0 GB	103.3 GB	104.4 GB	90.9 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Runs on Instinct MI250X 128 GB
118.2 GB	118.5 GB	119.6 GB	104.2 GB	3-bit · i-quant — Importance-weighted 3-bit	Runs on Instinct MI250X 128 GB
122.0 GB	122.3 GB	123.4 GB	98.6 GB	2-bit · small K-quant — 2-bit small K-quant	Runs on Instinct MI250X 128 GB
125.8 GB	126.1 GB	127.2 GB	118.2 GB	3-bit · i-quant extra small — Importance-weighted 3-bit, extra small	Runs on Instinct MI250X 128 GB
129.6 GB	129.9 GB	131 GB	116.1 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Runs on H200 NVL 141 GB
129.6 GB	129.9 GB	131 GB	117.3 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Runs on H200 NVL 141 GB
137.2 GB	137.5 GB	138.6 GB	135.8 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Runs on H200 NVL 141 GB
148.6 GB	149 GB	150 GB	128.1 GB	3-bit medium — 3-bit medium quantization	Runs on B200 192 GB
163.8 GB	164.2 GB	165.2 GB	136.7 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Runs on B200 192 GB
171.4 GB	171.8 GB	172.8 GB	136.7 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Runs on B200 192 GB
182.8 GB	183.2 GB	184.3 GB	~182.5 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Runs on B200 192 GB
217.1 GB	217.4 GB	218.5 GB	~216.7 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Runs on Instinct MI325X 256 GB
251.3 GB	251.6 GB	252.7 GB	~251 GB	6-bit, very good — 6-bit quantization, very good quality	Runs on Instinct MI325X 256 GB
304.5 GB	304.8 GB	305.9 GB	~304.2 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Runs on 5× H100 SXM 80 GB
608.7 GB	609 GB	610.1 GB	~608.4 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Runs on 5× H200 SXM 141 GB

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

183.0GBat 16K context

Weights: 182.5 GB
KV cache: 0.2 GB
Overhead: 0.3 GB

Server-class: ~183 GB at 16K context · one B200 192 GB

Or rent B300 from $6.94/hr.

Which Hardware Can Run DeepSeek V4 Flash 0731?

Q4_K_M · 182.8 GB

DeepSeek V4 Flash 0731 (Q4_K_M) requires 182.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 238+ GB is recommended. Using the full 1049K context window can add up to 11.5 GB, bringing total usage to 194.3 GB. No consumer GPU has enough memory. 4 devices with unified memory can run DeepSeek V4 Flash 0731, including Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 256GB).

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 ≥ 84.1 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →

How good is DeepSeek V4 Flash 0731?

The llmrun Score merges 21 public benchmark boards into a single figure, where 100 represents the average for open models.

Score vs VRAM in the ≤192 GB tier

DeepSeek V4 Flash 0731
Other models
Best Score for its VRAM
x: VRAM (GB, log scale) · y: llmrun Score

#2 of 78 that fit ≤192 GB

Each dot is a model at the best quant that fits ≤192 GB. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.

Full ≤192 GB ranking →

Benchmarks

GPQA Diamond
Epoch AI · Jul 2026Independent
91.0%
#4/98 open
Independent
LiveBench Coding
LiveBench · Jun 2026Independent
75.0
#10/19 open
Independent
LiveBench Reasoning
LiveBench · Jun 2026Independent
86.6
#4/19 open
Independent
AIME 2024/2025
Epoch AI · Jul 2026Independent
94.4%
#8/86 open
Independent
FrontierMath
Epoch AI · Jul 2026Independent
57.5%
#7/17 open
Independent
LiveBench Math
LiveBench · Jun 2026Independent
86.8
#10/19 open
Independent

All 21 DeepSeek V4 Flash 0731 benchmarks

In the ≤192 GB class

78 models fit ≤192 GB

Same Score, only models that fit ≤192 GB at their best quant: the ranking a ≤192 GB owner actually chooses from.

#1
MiMo V2.6 Flash RLXiaomi
131.9
~229 tok/s · 186.5 GB
#2
DeepSeek V4 Flash 0731DeepSeek
127.6
~181 tok/s · 183.0 GB
#3
GLM 5.3 FlashZ.ai
124.2
~228 tok/s · 185.1 GB
#4
Qwen3.8 27BAlibaba
122.1
~299 tok/s · 17.9 GB
#5
Inkling SmallThinking Machines
119.6
~399 tok/s · 162.7 GB
#6
Qwen3.5 397B A17BAlibaba
113.0
~185 tok/s · 182.1 GB

Full ≤192 GB ranking →B200 page →

Alternatives to DeepSeek V4 Flash 0731

DeepSeek V4 family · 12 models →

Official models with an llmrun Score, each answering one question; sizes are each model's default quant (usually Q4_K_M) with a 16K-token context.

Best alternative at this size
MiMo V2.6 Flash RL
Xiaomi
- 4.3 higher Score
- 3.5 GB more VRAM
Score 131.9 · 186.5 GB
Compare with DeepSeek V4 Flash 0731 Open MiMo V2.6 Flash RL
Other vendor
GLM 5.3 Flash
Z.ai
- 3.4 lower Score
- 22.2 GB more VRAM
Score 124.2 · 205.2 GB
Compare with DeepSeek V4 Flash 0731 Open GLM 5.3 Flash

Similar models

Model	Score	VRAM	tok/s	Why it is similar
Qwen3.8 27BAlibaba	122.1	17.9 GB	~38	similar score
Inkling SmallThinking Machines	119.6	162.7 GB	—needs > 24 GB	similar score
Qwen3.5 397B A17BAlibaba	113.0	242.6 GB	—needs > 24 GB	similar score
MiniMax M2.5MiniMax	111.8	139.6 GB	—needs > 24 GB	same size
GLM 4.7Z.ai	109.1	217.9 GB	—needs > 24 GB	same size
MiMo V2.5Xiaomi	108.3	187.6 GB	—needs > 24 GB	same size
Qwen3 235B A22B Thinking 2507Alibaba	108.2	142.9 GB	—needs > 24 GB	same size
MiniMax M2.7MiniMax	107.4	139.6 GB	—needs > 24 GB	same size
MiniMax M2.1MiniMax	106.2	139.6 GB	—needs > 24 GB	same size
MiniMax M2MiniMax	104.0	139.6 GB	—needs > 24 GB	same size

Fine-tunes & variants(6)community builds of DeepSeek V4 Flash 0731

Specifications

Publisher: DeepSeek
Family: DeepSeek V4
Parameters: 304.2B
Active parameters: 27.4B per token
Architecture: DeepseekV4ForCausalLM
Context Length: 1,048,576 tokens
Vocabulary Size: 129,280
Release Date: 2026-07-31
License: MIT

Resources

Frequently Asked Questions

How much VRAM does DeepSeek V4 Flash 0731 need?

DeepSeek V4 Flash 0731 requires 182.8 GB of VRAM at Q4_K_M, or 608.7 GB at BF16. Full 1049K context adds up to 11.5 GB (194.3 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 304.2B × 4.8 bits ÷ 8 = 182.5 GB

KV Cache + Overhead ≈ 0.3 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 ≈ 11.8 GB (at full 1049K context)

VRAM usage by quantization

Q4_K_M

182.8 GB

Q4_K_M + full context

194.3 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run DeepSeek V4 Flash 0731?

No — DeepSeek V4 Flash 0731 requires at least 84.0 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for DeepSeek V4 Flash 0731?

For DeepSeek V4 Flash 0731, Q4_K_M (182.8 GB) offers the best balance of quality and VRAM usage. Q5_K_M (217.1 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 84.0 GB.

VRAM requirement by quantization

IQ2_XXS

84.0 GB~53%

IQ3_XXS

118.2 GB~70%

IQ3_M

137.2 GB~78%

Q4_K_M ★

182.8 GB~89%

Q5_K_M

217.1 GB~92%

BF16

608.7 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run DeepSeek V4 Flash 0731 on a Mac?

Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run DeepSeek V4 Flash 0731, the smallest at IQ2_XS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 182.8 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (192 GB).

Can I run DeepSeek V4 Flash 0731 locally?

Yes — DeepSeek V4 Flash 0731 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is DeepSeek V4 Flash 0731?

At Q4_K_M, DeepSeek V4 Flash 0731 can reach ~83 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 → 16.4 GB active ÷ (8000 × 0.65) = 3.16 ms, plus 43 layers × 0.055 ms = 2.37 ms, so 1000 ÷ 5.53 ms = ~181 tok/s

Estimated speed at Q4_K_M (182.8 GB)

NVIDIA B200

~181 tok/s

NVIDIA B300

~181 tok/s

AMD Instinct MI350X

~83 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 V4 Flash 0731?

At Q4_K_M, the download is about 182.51 GB. The full-precision BF16 version is 608.36 GB. The smallest option (IQ2_XXS) is 83.65 GB.

Which GPUs can run DeepSeek V4 Flash 0731?

No single consumer GPU has enough VRAM to run DeepSeek V4 Flash 0731 at Q4_K_M (182.8 GB). Multi-GPU or professional hardware is required.

Which devices can run DeepSeek V4 Flash 0731?

4 devices with unified memory can run DeepSeek V4 Flash 0731 at Q4_K_M (182.8 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.

DeepSeek V4 Flash 0731 — Hardware Requirements & GPU Compatibility

DeepSeek V4 Flash 0731 runs on a 192 GB+ Mac or a datacenter GPU

llmrun Composite Score →

Run DeepSeek V4 Flash 0731

Recommended

Other ways

Run in cloud

How Much VRAM Does DeepSeek V4 Flash 0731 Need?

Selected Q4_K_M

Which Hardware Can Run DeepSeek V4 Flash 0731?

How good is DeepSeek V4 Flash 0731?

Score vs VRAM in the ≤192 GB tier

Benchmarks

In the ≤192 GB class

Alternatives to DeepSeek V4 Flash 0731

MiMo V2.6 Flash RL

GLM 5.3 Flash

Similar models

Specifications

Resources

Frequently Asked Questions