DeepSeek V4.1 Is Here, Plus GPT Image 2.5 for Image Generation

DeepSeek V4.1 Flash is here: a million-token context window, 384K output, and native vision and tool calling. This release also brings in the GPT Image 2.5 models, so AI experts can generate and edit images directly.

## Four DeepSeek V4.1 Flash variants

Pick based on how hard the job is:

| Model | Thinking | Best for |
|------|------|------|
| DeepSeek V4.1 Flash | Off | Everyday chat, quick tasks |
| DeepSeek V4.1 Flash Thinking Low | Low | Light reasoning, saves tokens |
| DeepSeek V4.1 Flash Thinking High | High | Everyday reasoning |
| DeepSeek V4.1 Flash Thinking Max | Max | Complex agent tasks |

All four share the same specs: **1M context** (1,048,576 tokens) and **up to 384K output tokens**, with vision, tool calling, streaming, and JSON support.

Both the DeepSeek and OpenCode Go providers are updated — just pick one when you configure.

## Image generation: GPT Image 2.5

Two new image models:

| Model | Capabilities |
|------|------|
| GPT Image 2.5 Flare | Image generation + editing |
| GPT Image 2.5 Sunburst | Image generation + editing |

Once enabled, AI experts can generate images from a description. Editing uses multipart upload to modify an existing image.

## Other models in this release

| Model | Context |
|------|------|
| GLM-5.3 / GLM-5.3-Flash | 1M |
| Qwen3.8 Flash | 1M |
| LongCat 2.0 | 1M |
| Hy4 Preview | 960K |

## How to use

Settings → Model configuration → pick a provider, then enable the models you want.

## Requirements

- Yao **1.0.0-rc21+**

Grab the latest version from the [download page](/download), or check for updates in Yao Engine Desktop.