whatmodel.sh / DeepSeek V4 Flash Vision (Exp)

DeepSeek V4 Flash Vision (Exp)

+ contribute

DeepSeek V4 Flash Vision (Exp) is an open-weights language model from DeepSeek, released August 2026. Its published size is 304.6B parameters and a 1M-token context window. Licence: mit.

Catalogue identity notes

DeepSeek's 'first experimental multimodal model in the DeepSeek-V4 family. It builds on the DeepSeek-V4-Flash architecture by incorporating visual modules and undergoing continued training' (repo @6821d6a, created 2026-08-31, MIT). A separate checkpoint from DeepSeek-V4-Flash-0731 (catalogued deepseek-v4-flash), which the card compares against. No first-party API id found (DeepSeek API docs not checked for a vision alias).

Model information sources

Release date ↗ · Weights licence ↗

Safetensors metadata total 304,646,824,126 parameters (repo @6821d6a, includes the vision encoder/aligner and MTP/DSpark weights); the card states no parameter count or active parameters, so active_b is left out.

75.7±2.3#35 of 719 LLMsfrom 8 benchmark results across 2 sources

$ how it compares

The LLMs ranked either side of DeepSeek V4 Flash Vision (Exp), out of 828 with a score, then more from DeepSeek.

#	model	score	where	RAM	released
34	Qwen3.8 Max (0902) Alibaba	76±1	api $3.00/M		2026-09
35	DeepSeek V3.2 Speciale DeepSeek	76±3	comparison only
36	DeepSeek V4 Flash Vision (Exp) DeepSeek	76±2	local	~104GB	2026-08
37	Qwen3.8-Flash-Next Alibaba	75±1	local	~59GB	2026-08
38	Qwen3.8-Omni-Flash Alibaba	75±2	api $0.23/M		2026-09
	more from DeepSeek
21	DeepSeek V4.1 Flash DeepSeek	78±1	api $0.26/M +local	~537GB	2026-09
31	DeepSeek V4 Pro (0813) DeepSeek	76±1	api $1.63/M +local	~1010GB	2026-08
50	DeepSeek V4 Flash (0731) DeepSeek	74±1	api $0.11/M +local	~92GB	2026-07

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #32 of 300 · 57% of its score · 4 tests

Beats GPT-5.4, GPT-6 Luna, GPT-5.6 Luna here, though they rank higher overall. Trails GLM-5.3-Flash, Grok 4.7 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Terminal-Bench 2.1 lab	83.9% ✎	▁·▁·▁▁▄▃▄█▃▇ beats 43/56	hard · avg 22%	▇▇▇▇▇▇▇▇▇▇ 25%	small run
DeepSWE v1.1 (resolved) lab	59.3% ✎	▂····▁▁▁▆██▆ beats 8/28	hard · avg 19%	▇▇▇▇▇ 12%	at its level · small run
Agents' Last Exam (metric/harness unspecified) lab	27.3% ✎	▂·▂·█▂···▂▄▄ beats 4/11	very hard · avg 7.1%	▇▇▇▇ 11%	at its level · small run
Toolathlon-Verified lab	75.9% ✎	▂·▂▂▅▃···▂██ beats 12/18	hard · avg 33%	▇▇▇ 8.4%	blunt · at its level · small run

coding▁▁▁▃▆██▇▆▄▃▁ #72 of 458 * · 12% of its score · 1 test

Trails Gemini 3.5 Flash-Lite, MiMo-V2.5 (310B-A15B), MiMo-V2.5-Pro (1T-A42B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
DeepSWE v1.1 (resolved) lab	59.3% ✎	▂····▁▁▁▆██▆ beats 8/28	hard · avg 19%	▇▇▇▇▇ 12%	at its level · small run

vision & documents▁▁▃▂▃▃▄▂▆▇█▄ #121 of 147 * · 0.8% of its score · 1 test

Trails Qwen3.5 2B, Qwen3-VL 2B Thinking, Qwen2.5-VL 7B Instruct here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
PureDocBench document parsing (Avg3 over Clean/Digital/Real)	50.1% ✎	▃·▁·▂▂▅█·▁▂▂ beats 0/22	medium · avg 62%	▇ 0.8%	blunt · at its level

other tests3 not in a task · 40% of its score

test	result	vs the others in its run	difficulty	counts	why
NL2Repo lab	57.7% ✎	▂▂·▂▂▄█▆▆▂▄▆ beats 18/21	very hard · avg 12%	▇▇▇▇▇▇▇ 17%	at its level · small run
CyberGym lab	75.3% ✎	▄···▄··▄█▄█▄ beats 3/8	hard · avg 31%	▇▇▇▇▇ 12%	at its level · small run
AutomationBench (Public) lab	25.7% ✎	█··········█ beats 1/1	very hard · avg 6.2%	▇▇▇▇ 10%	at its level · small run

$ where to run it

llama.cpp · recommended

version

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run DeepSeek V4 Flash Vision (Exp) · needs ~156.03 GB · from huggingface.co/unsloth
chat in the terminal
llama cli -hf unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF:UD-Q4_K_XL
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF:UD-Q4_K_XL

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run DeepSeek V4 Flash Vision (Exp) · needs ~156.03 GB · from huggingface.co/unsloth
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/unsloth/DeepSeek-V4-Flash-Vision-Exp-GGUF@UD-Q4_K_XL -y
Then run lms ls to see its name, and lms chat followed by that name

Other ways to run it

Run it as published · needs ~167.82 GB, NVIDIA only · from huggingface.co/deepseek-ai
sglang serve --model-path deepseek-ai/DeepSeek-V4-Flash-Vision-Exp --tp 4 --speculative-algorithm DSPARK

Weights and documentation: huggingface.co/deepseek-ai ↗

$ how the score was worked out · deepseek-ai/DeepSeek-V4-Flash… 97%, How Far Is Document Parsing f… 1%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
deepseek-ai/DeepSeek-V4-Flash-Vision-Exp model card · lab-reported	7	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 97%	MIT
How Far Is Document Parsing from Solved? PureDocBench: A Source-Traceable Benchmark across Clean, Degraded, and Real-World Settings (arXiv:2605.07492v2)	1	▇ 1%	CC BY 4.0
prior · pull toward the middle		▇ 2%
weighted average		= 75.7

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-10-11 02:20 UTC · sources and licences · json