DeepSeek V4 Pro 0813 and V4 Flash Vision Exp are available in iolys
DeepSeek V4 Pro 0813 and Flash Vision Exp are now natively available in iolys for .NET developers, with a 1M-token context window and vision.
The DeepSeek V4 family is expanding with two complementary models: DeepSeek V4 Pro 0813, built for the most demanding tasks, and DeepSeek V4 Flash Vision Exp, which adds image understanding to the fast and cost-efficient branch of the lineup.
Read DeepSeek's official announcements for V4 Pro 0813 and V4 Flash Vision Exp for the release details and published results.
Both models are now natively supported by iolys for C# and .NET developers inside Visual Studio. You can select them in the same environment as your active solution without relying on a separate tool.
With iolys, .NET developers get native DeepSeek integration in Visual Studio: V4 Pro for complex tasks, V4 Flash for speed, and Vision Exp when images are part of the context.
DeepSeek V4 Pro 0813 for the most complex tasks
Released on August 13, 2026, DeepSeek V4 Pro 0813 is the official version of V4 Pro. It replaces the preview with stronger agentic capabilities and improvements that DeepSeek describes as particularly visible in production environments.
V4 Pro targets situations where depth of analysis matters most: understanding a large solution, preparing a cross-cutting change, reasoning about an architecture, coordinating several tools, or solving a difficult development problem.
This release also offers:
- three reasoning-effort levels:
low,high, andmax; - native support for the OpenAI Responses API;
- a context window of up to one million tokens;
- reported improvements across several agent and software-development benchmarks.
The official DeepSeek V4 Pro 0813 model card provides details about the model and the results published by DeepSeek.
V4 Flash Vision Exp brings vision to the Flash family
The second model is officially named DeepSeek V4 Flash Vision Exp. The Exp suffix indicates that it is still an experimental model.
Vision Exp builds on the V4 Flash architecture, to which DeepSeek added visual modules and continued training focused on image understanding. It can therefore combine text and images in the same request.
In practical terms, an agent can now use:
- an application screenshot;
- a user-interface mockup;
- an architecture diagram;
- a chart or table contained in an image;
- an error visible in Visual Studio;
- scanned documentation or specifications.
This is not a simple switch that turns V4 Flash 0731 into a visual model. Vision Exp is a distinct model, identified as deepseek-v4-flash-vision-exp, and must be selected when sending images. It nevertheless retains text performance reported as comparable to V4 Flash while substantially improving tasks that require visual understanding.
The API accepts requests combining text and images through Chat Completions, Messages, or Responses. Images can be provided as base64, by URL, or through the Files API. DeepSeek applies V4 Flash pricing to Vision Exp.
More details are available in the official DeepSeek V4 Flash Vision Exp announcement and the Vision API guide.
A one-million-token context window across the lineup
V4 Pro 0813, V4 Flash 0731, and V4 Flash Vision Exp all provide a context window of up to one million tokens. This is particularly valuable in software development, where a request may depend on many files, a long conversation, specifications, and results produced by several tools.
In iolys, this extended context lets the agent retain more useful information around the active Visual Studio solution. With Vision Exp, images become part of that context: a screenshot or diagram can be analyzed together with the instructions, code, and session history.
Such a high limit does not mean that the entire window should always be filled. Targeted context remains faster and more precise. The one-million-token capacity primarily provides comfortable headroom for large projects and long-running tasks.
Very low prices, including for vision
Since August 16, 2026, DeepSeek has applied different peak and off-peak prices. As of August 22, the listed prices per million tokens are:
| Model | Uncached input | Output |
|---|---|---|
| V4 Flash 0731 and Vision Exp, off-peak | $0.22 | $0.66 |
| V4 Flash 0731 and Vision Exp, peak | $0.44 | $1.32 |
| V4 Pro 0813, off-peak | $0.66 | $1.98 |
| V4 Pro 0813, peak | $1.32 | $3.96 |
Cached input costs even less: starting at $0.007 per million tokens for Flash and Vision Exp, and $0.022 for Pro during off-peak hours. Images sent to Vision Exp are converted into tokens and billed at the V4 Flash rate.
Even V4 Pro, the most ambitious model in the lineup, therefore retains highly aggressive pricing. Flash and Vision Exp are particularly attractive for frequent agent interactions, regular screenshot analysis, or work with a large context window without allowing the cost of every request to become the primary decision factor.
Schedules and prices may change, so the official DeepSeek pricing page remains the reference when estimating usage volume.
Two provider options in iolys
For a direct connection, the DeepSeek Platform lets you fund your account and create the API key used by iolys. Calls to V4 Pro 0813, V4 Flash 0731, or V4 Flash Vision Exp are then billed directly by DeepSeek.
The V4 family is also available through OpenRouter. With an OpenRouter account and API key, you can select:
deepseek/deepseek-v4-pro-0813for V4 Pro;deepseek/deepseek-v4-flash-0731for V4 Flash;deepseek/deepseek-v4-flash-vision-expfor tasks involving images.
OpenRouter aggregates multiple inference providers and may route requests based on availability, speed, or price. Its pricing conditions may therefore differ from those of the direct DeepSeek API.
OpenRouter is also natively supported by iolys. Configure the provider and its API key without adding an external connector. You can choose between a direct relationship with DeepSeek and a multi-provider catalog while keeping the same Visual Studio workspace.
Native DeepSeek integration for .NET developers
With iolys, C# and .NET developers can use these new models without leaving their Visual Studio workflow:
- choose V4 Pro 0813 for the most complex reasoning and development tasks;
- choose V4 Flash Vision Exp when a screenshot, diagram, or other image must be part of the context;
- return to V4 Flash 0731 when speed and cost are the main priorities.
DeepSeek support is native in iolys: no intermediary extension or custom model configuration is required. Keep your DeepSeek connection, active C# or .NET solution, and iolys tools, then simply select the model that best fits the task.
Use DeepSeek in Visual Studio or configure OpenRouter in iolys
Sources: official V4 Pro announcement from August 13, 2026, official Vision Exp announcement from August 21, 2026, and the DeepSeek release notes.