TLDR
- DeepSeek launched V4-Flash-Vision-Exp, adding image and screenshot understanding to its existing V4 Flash model.
- DeepSeek says its experimental model approaches Claude Opus 4.8 performance on multimodal agent benchmarks.
- The experimental model retains V4 Flash capabilities across text reasoning, agent tasks and world knowledge.
- DeepSeek made its new multimodal V4 Flash Vision model available through the company’s API platform.
- DeepSeek’s latest release expands competition with Anthropic across multimodal and autonomous AI agent workloads globally.
DeepSeek has released an experimental multimodal version of its V4 Flash artificial intelligence model, adding image understanding while claiming performance close to Anthropic’s Claude Opus 4.8 on agent benchmarks.
DeepSeek V4 Flash Adds Image Understanding
The Chinese AI developer launched DeepSeek-V4-Flash-Vision-Exp on August 21 as an experimental model available through its API. The system expands the existing V4 Flash model by allowing users to submit images and text together.
DeepSeek said the model can interpret visual material such as images and screenshots while using that information during agent-based tasks. Its text performance remains comparable with the standard V4 Flash model across reasoning, agent tasks and general knowledge, according to the company
The company claims V4-Flash-Vision-Exp approaches Anthropic’s Claude Opus 4.8 on benchmarks requiring both visual understanding and agent capabilities. The comparison remains a company-reported benchmark rather than proof that the models perform equally across workloads.
Images submitted through the API are converted into tokens, with each image using no more than 384 tokens after processing. DeepSeek is also supporting image uploads through base64 data, external URLs and its Files API.
DeepSeek Targets Lower AI Inference Costs
The experimental release builds on DeepSeek’s V4 family, which entered preview in April with V4 Pro and V4 Flash. Both models support context windows of up to one million tokens, while V4 Flash was designed as the smaller and cheaper option.
DeepSeek later released V4-Flash-0731, which improved agent performance while retaining the same model structure. Published benchmark results put V4 Flash close to Opus 4.8 on some coding and agent tests, although Anthropic’s model remained ahead across DeepSeek’s reported comparison table.
Price remains a major difference between the models. DeepSeek V4 Flash has been priced at $0.28 per million output tokens, compared with $25 per million output tokens for Claude Opus 4.8 in published pricing comparisons. That is a token-pricing comparison, rather than $0.28 versus $25 for every completed response, since the final cost depends on how many tokens a task consumes.
Chinese AI Developers Increase Pressure on US Rivals
DeepSeek’s latest release adds another multimodal model to a market where Chinese AI companies are competing with U.S. developers on performance, pricing and developer access. The company has focused heavily on reducing the computing resources required to run its models while maintaining competitive benchmark results.
The V4 architecture uses techniques designed to reduce memory and computing requirements during long-context processing. DeepSeek also supports OpenAI Chat Completions and Anthropic-compatible API formats, allowing developers already using those interfaces to integrate its models with fewer changes.
The company has followed the April V4 preview with several updates as competition around coding agents and multimodal systems increases. V4-Flash-Vision-Exp remains experimental, meaning its current performance and features could change before DeepSeek moves the vision model into a production release.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
