Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks

Chinese AI company Deepseek released Deepseek-V4-Flash-Vision-Exp, an experimental multimodal model that adds image understanding to its text capabilities, and on Deepseek's internal benchmarks it nearly matches Opus 4.8 on agent tasks. The model supports JPEG, PNG, GIF, and WebP, works with OpenAI's Chat Completions and Responses APIs and Anthropic's Messages endpoint, and is priced at V4-Flash rates with each image costing at most 384 tokens. Deepseek also released version 0.1.1 of its Harness framework, which supports the new model out of the box.

Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks Chinese AI company Deepseek has released V4-Flash-Vision-Exp, an experimental multimodal model that adds image understanding to its text capabilities. On Deepseek's own benchmarks, the model nearly matches Opus 4.8 on agent tasks. Deepseek-V4-Flash-Vision-Exp https://api-docs.deepseek.com/guides/vision/ extends Deepseek-V4-Flash https://the-decoder.com/new-deepseek-flash-model-matches-openais-gpt-5-6-luna-at-roughly-60-percent-lower-cost/ with image processing while keeping the base model's text performance in reasoning and world knowledge, Deepseek says https://x.com/deepseek ai/status/2090730032574631962 . On the company's internal multimodal agent benchmarks, the vision variant scores close to Opus 4.8. Deepseek is targeting visual agent workflows Deepseek is positioning the model for agent-based applications. It's designed to work with different agent frameworks and combine visual understanding with tool use. In practice, it can describe images, extract text from screenshots, and analyze diagrams. It handles JPEG, PNG, GIF, and WebP, and determines the format from actual file content rather than the filename or declared MIME type, per the API docs https://api-docs.deepseek.com/guides/vision/ . The model works with OpenAI's Chat Completions and Responses APIs and Anthropic's Messages endpoint. Deepseek also released version 0.1.1 of its Harness framework https://the-decoder.com/deepseek-launches-an-improved-v4-pro-model-raises-api-prices-and-makes-its-agent-software-open-source/ , which supports the new model out of the box. Pricing and image limits There are three ways to send images to the model. Developers can embed them directly with Base64 encoding, point to publicly accessible URLs up to 32 MiB , or use the new, free Files API https://api-docs.deepseek.com/guides/files api . The Files API lets you upload a file once and reference it by ID across multiple requests, with a size limit of 64 MiB. An optional "detail" field downscales images to 512 x 512 pixels, saving tokens when fine visual detail isn't needed. The model automatically normalizes images to roughly 800 x 800 pixels depending on the aspect ratio before processing. Regardless of original resolution, each image costs at most 384 tokens. Pricing follows V4-Flash rates. A single request can include up to 600 images. Max edge length is 8,192 pixels per side, but that drops to 4,096 pixels once a request contains 15 or more images. Images can only go in user messages. AI News Without the Hype – Curated by Humans Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section. Subscribe now Deepseek https://x.com/deepseek ai/status/2090730032574631962