DeepSeek · Model guide
DeepSeek V4 Flash Vision Exp
An experimental multimodal member of the DeepSeek V4 family: it adds image understanding on top of V4-Flash while matching it on text agents, reasoning and world knowledge. Text and images can be interleaved in user messages, the 1M-token context window is unchanged, and it targets agents that have to look at a screenshot, chart or document before acting.
What this model is good at
What the vendor positions it for, and which of our models to reach for instead.
VendorDeepSeek positions it as an experimental multimodal model: on par with V4-Flash for pure text, with a large jump on vision-dependent agent benchmarks that brings multimodal agent capability close to Opus-4.8.source
Vision-dependent agents · The vendor reports a large jump over V4-Flash on agent benchmarks that require vision, approaching Opus-4.8.(vendor claim)source
Charts and document images · Scores 64.3 on Chartography and takes up to 600 images per request via base64, URL or the Files API.source
Coding agents that read screenshots · Text capability is on par with V4-Flash, so vision is added without giving up the text agent scores.(vendor claim)source
- deepseek/deepseek-v4-flashsibling —Choose V4-Flash for pure-text workloads or when an experimental model is not acceptable.(vendor claim)source
- deepseek/deepseek-v4-prostronger —Choose Pro for the hardest knowledge, research and complex-agent tasks — it has no image input.(vendor claim)source
Pricing and billing
Billed per token. Cached input is charged at the cache-read rate.
| Price / 1M tokens | You pay |
|---|---|
| Input | $ 0.45$ 0.41 |
| Output | $ 1.35$ 1.22 |
| Cache read | $ 0.015$ 0.013 |
| Cache write | $ 0.45$ 0.41 |
+ 2.0K × $ 1.22
Capabilities and limits
What CrossModel guarantees across every route this model can take right now.
- Context window
- 1.0M tokens
- Max output
- 384.0K tokens
- Input / output modalities
- Text + Image → Text
- Streaming
- Supported
- Tool calling
- Supported
- Structured output (JSON)
- Supported
- Image input
- Supported
- Thinking mode
- Supported· can be turned off
- Reasoning effort
- low · medium · high · xhigh · max
- Available endpoints
- /v1/chat/completions · /v1/responses · /v1/messages
Published benchmarks
Scores the vendor reported, with the evaluation setup each one came from.
- Agents' Last Examscore·DeepSeek · 2026-08-21setup not fully disclosed
厂商自测。该测评含多模态元素，纯文本的 V4-Flash 会忽略这些元素；来源页未披露该项的评测设置。
27.3% - ApexBenchpass_at_1·DeepSeek · 2026-08-21setup not fully disclosed
厂商自测，Pass@1。该测评含多模态元素，纯文本的 V4-Flash 会忽略这些元素，所以与 V4-Flash 的同名成绩不是等价对比。名称与目录里 x-ai 的 APEX-Agents / APEX-SWE 是否同一测评未获证实，故单独立项。
36.5% - AutomationBench (Public)score·DeepSeek · 2026-08-21setup not fully disclosed
厂商自测的 Public 划分；与不带 (Public) 的 AutomationBench 不是同一身份，不要混用。来源页未披露该项的评测设置。
25.7%
- DeepSWEresolved_rate·DeepSeek · 2026-08-21effort maxDeepSeek Harness minimalsetup not fully disclosed
厂商自测，Code Agent 文本任务口径；来源页未标注 DeepSWE 版本号。
59.3% - DSBench-Hardscore·DeepSeek · 2026-08-21setup not fully disclosed
DeepSeek 内部使用的 Coding Agent 难题测试集，不是公开基准，无法交叉验证；来源页未披露该项的评测设置。
63.6% - NL2Reposcore·DeepSeek · 2026-08-21effort maxDeepSeek Harness minimalsetup not fully disclosed
厂商自测，Code Agent 文本任务口径；来源页只给出百分数，未说明具体 metric，保守记为 score。
57.7%
Vendor-reported numbers, not CrossModel measurements. Scores are only comparable when the benchmark version, metric and evaluation setup match, so nothing here is averaged or ranked.
Use it in your tools
Point the base URL at CrossModel and paste this model ID — every tool below has a setup guide.
Related models
Other tiers from the same vendor, plus similarly priced alternatives.
Frequently asked questions
What is DeepSeek V4 Flash Vision Exp?
How much does DeepSeek V4 Flash Vision Exp cost?
Does DeepSeek V4 Flash Vision Exp support tool calling and structured output?
Which endpoint do I call?
Can I try it without writing code?
Chat first, integrate later
Chat first, then wire it in once you like the answers.