DeepSeek-V4-Flash-Vision-Exp

Experimental multimodal vision-understanding sibling of DeepSeek-V4-Flash, and one of the shortest-lived endpoints in this catalog: released 2026-08-21 and retired 2026-09-10, twenty days later. Both dates are DeepSeek's own, from its Change Log — 'Today, the new multimodal vision understanding model DeepSeek-V4-Flash-Vision-Exp is now available on the DeepSeek API platform. This is an experimental model that can be accessed by setting model=deepseek-v4-flash-vision-exp' (2026-08-21), and 'The previous-generation models V4 Flash and V4 Flash Vision Exp have been retired; for compatibility, the model names deepseek-v4-flash and deepseek-v4-flash-vision-exp are temporarily routed to V4.1 Flash' (2026-09-10). RETIREMENT SIGNAL: an explicit past-tense statement by the provider, on a dated announcement — not an inference from the model disappearing off the pricing page, though it did also disappear from it the same day. The name is still ACCEPTED by the API, so calls do not fail; they are served by DeepSeek-V4.1-Flash and billed at the Flash price, and DeepSeek calls that routing 'temporary'. On DeepSeek's own benchmarks it was on par with V4-Flash on pure text and a significant step up on visual agent tasks. PRICES ABOVE ARE THE RATES IN FORCE AT RETIREMENT and deliberately stop tracking anything: PEAK per 1M was $0.014 cache-hit / $0.44 cache-miss / $1.32 out, off-peak exactly half ($0.007 / $0.22 / $0.66), identical to deepseek-v4-flash on every axis, with images converted to input tokens by dimension. Those figures come from DeepSeek's pricing page as it stood on 2026-09-07, read from the Internet Archive snapshot cited below, because the live page dropped the column on the day of retirement — this row was added on 2026-09-10, after the fact, so the archived first-party document is the only surface that still carries them. Context 1M, max output 384K, concurrency limit 2500, JSON output / tool calls / Responses API / Anthropic API / chat prefix completion supported, FIM completion NOT supported (it was the only V4 endpoint where FIM was unavailable). Knowledge cutoff never published. Open weights at huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp.

Source: DeepSeek official documentation ↗

Explore more

DeepSeek-V4-Flash-Vision-Exp — questions & answers