← All models

DeepSeek-V4-Flash-Vision-Exp

Published Publisher DeepSeekdeepseek-v4-flash-vision-exp

Overview

Released on August 21, 2026, DeepSeek-V4-Flash-Vision-Exp is an experimental multimodal vision-understanding model built on DeepSeek-V4-Flash. It accepts mixed text and image input and is intended for visual understanding and tool-assisted agent workflows. DeepSeek reports text-only agent, reasoning and world-knowledge capabilities comparable to V4-Flash, with improved performance on visual agent benchmarks. (api-docs.deepseek.com)

Capabilities and uses

At release, the API supported Chat Completions, Messages and Responses, with images supplied by base64, external URLs or the Files API. Potential uses include image description, reading screenshots, chart analysis and multimodal agent workflows. The publisher also provides model weights and inference materials under an MIT license. (api-docs.deepseek.com)

Availability and limitations

  • The original model has been retired from DeepSeek's API. As of November 16, 2026, the legacy ID remains accepted temporarily, but requests are served by DeepSeek-V4.1-Flash-1017 and billed at its Flash rates. The original model and the currently routed model should not be treated as identical. (api-docs.deepseek.com)
  • The release announcement described the model as experimental and stated that images were billed at up to 384 tokens each. The current vision guide describes different image-token rules for the replacement Flash model, including a maximum of 1,024 tokens per image; do not assume the historical limit applies to current routed requests. (api-docs.deepseek.com)
  • The official launch announcement did not specify numeric historical input and output token prices; those historical amounts were not verified from the official sources consulted. (api-docs.deepseek.com)

Current API pricing for the legacy ID

The following are replacement-model rates, not verified historical prices for the original model. All amounts are per 1 million tokens. Prices were checked on November 16, 2026.

Official price listing Period Input, cache hit Input, cache miss Output
International, USD Off-peak $0.003 $0.15 $0.60
International, USD Peak $0.006 $0.30 $1.20
Mainland China, CNY Off-peak ¥0.02 ¥1 ¥4
Mainland China, CNY Peak ¥0.04 ¥2 ¥8

Peak hours are Monday–Friday, 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays; all other times are off-peak. The two currency listings are official regional listings, not currency conversions. (api-docs.deepseek.com)

Official sources: Release announcement; international USD pricing; mainland CNY pricing; vision guide; publisher's model repository.