DeepSeek Releases V4-Flash-Vision-Exp Multimodal Model
DeepSeek launched an experimental multimodal model, V4-Flash-Vision-Exp, on its API platform along with a new free Files API.
ひとことで言うと
What did DeepSeek announce with the V4-Flash-Vision-Exp release?
DeepSeek released DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal model on its API platform. The company says it matches DeepSeek-V4-Flash on text tasks while adding vision, and claims a major jump in multimodal agent benchmark performance over V4-Flash, near Opus-4.8. DeepSeek also launched a free Files API and DeepSeek Harness 0.1.1.
要点
- DeepSeek released DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal model, on the DeepSeek API Platform on August 21, 2026.
- DeepSeek says the model matches DeepSeek-V4-Flash on text capabilities, including agents, reasoning, and world knowledge, while adding image understanding.
- DeepSeek says the model shows a major improvement over V4-Flash on multimodal agent benchmarks, bringing performance close to Opus-4.8, according to the company.
- DeepSeek launched a free Files API that lets developers upload an image once and reuse it across requests by referencing a file_id.
- DeepSeek also released DeepSeek Harness 0.1.1 the same day, which the company says provides out-of-the-box support for the new model.
DeepSeek releases an experimental multimodal model
DeepSeek has released DeepSeek-V4-Flash-Vision-Exp on the DeepSeek API Platform, an experimental model that adds multimodal capabilities to its V4-Flash line. Developers can access it by setting model='deepseek-v4-flash-vision-exp' in their API calls.
Text and multimodal performance claims
DeepSeek says the new model matches DeepSeek-V4-Flash on text capabilities, including agents, reasoning, and world knowledge. The company also says that on multimodal agent benchmarks, V4-Flash-Vision-Exp shows a major improvement over V4-Flash, and that this brings its multimodal agent performance close to Opus-4.8, according to DeepSeek. The announcement does not include specific benchmark scores.
DeepSeek says the model is built to work across agent frameworks, combining visual understanding with a range of tools, which the company describes as unlocking more practical agent workflows.
Multimodal API support
DeepSeek points developers to its Vision API guide for further documentation on using the multimodal endpoints.
Files API launched
Alongside the model, DeepSeek launched a Files API. The company says it is free to use and lets developers upload an image once, then reference it in later requests by its file_id rather than re-uploading it, which DeepSeek says saves request bandwidth. Uploaded images can be reused across multiple requests without re-uploading.
Harness update
DeepSeek also released DeepSeek Harness 0.1.1 on the same day. The company says this release provides out-of-the-box support for the new V4-Flash-Vision-Exp model.
Source: DeepSeek, "DeepSeek-V4-Flash-Vision-Exp Release: Multimodal API Now Live," DeepSeek API Docs, August 21, 2026.
よくある質問
- What model did DeepSeek release?
- DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal version of DeepSeek-V4-Flash, now available on the DeepSeek API Platform.
- How is image input billed?
- DeepSeek says images are tokenized for billing at up to 384 tokens each, charged at DeepSeek-V4-Flash pricing.
- What is the new Files API?
- A free feature that lets developers upload an image once and reference it by file_id in later requests instead of re-uploading it, according to DeepSeek.
- Which API formats does the model support?
- DeepSeek says the model supports Chat Completions, Messages, and Responses, with mixed text and image input via base64, external URLs, or the Files API.