DeepSeek V4 Flash Vision ExpPreview

DeepSeek · 304.6B parameters · Language models

Weights and memory

1 option
FormatFile sizeYour deviceLink
—168 GB158 GB ◇Hugging Face

Published by the lab. File size is what the repository lists; for original weights it is the total of the weight files, which can include more than one variant. ● memory stated by the publisher · ◇ estimate from file size, an 8K-token context and 0.5 GB for the runtime. Unknown means there is no figure to go on. Checked 2026-10-11.

Tests on this site

Not tested yet

This site has not run this model. What will be tested

Third-party scores

Measured by other sites on hosted endpoints. A quantized file you run yourself may score differently.

OpenRouter evals

data from 2026-10-10
EndpointBenchmarkScoreTasks run
deepseek/deepseek-v4-flash-vision-exp-2026082188.0% ± 1.11,188
deepseek/deepseek-v4-flash-vision-exp-2026082174.8% ± 2.0250

Source: OpenRouter evals (openrouter.ai) via OpenRouter (openrouter.ai/rankings). openrouter.ai

Artificial Analysis

data from 2026-10-09
EndpointIntelligenceCodingAgentic
DeepSeek V4 Flash Vision (Max)34.865.047.5

Source: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings). artificialanalysis.ai