DeepSeek V4 Flash Vision Exp Cuts API Prices 50% Across Input and Output

What Changed

DeepSeek has cut pricing on V4 Flash Vision Exp by exactly -50% across the board. Input tokens drop from $0.44 to $0.22 per 1M tokens; output tokens fall from $1.32 to $0.66 per 1M tokens. No partial discounts, no tier conditions — both sides of the pricing ledger are halved.

Why It Matters

For vision-heavy workloads — document parsing, image captioning, multimodal pipelines — output costs tend to dominate the bill. Moving from $1.32 to $0.66 per 1M output tokens is a meaningful shift, not rounding noise. A pipeline burning 10M output tokens monthly goes from $13.20 to $6.60. That compounds fast at scale.

The input cut is less dramatic in absolute terms (from $0.44 to $0.22), but matters for applications that push large image batches or long system prompts into the model repeatedly.

Who Should Care

• Developers running vision inference at volume — the math just got noticeably friendlier.
• Teams evaluating multimodal models — this brings V4 Flash Vision Exp closer to commodity pricing territory, making it easier to justify experimentation without budget approval.
• Anyone comparing against alternatives — check DeepSeek V4 Flash Vision Exp — live specs & price history for the full context before finalising a model selection.

Caveats

Price alone doesn't determine fit. Benchmark your actual tasks — latency, accuracy on your specific image types, and context window behaviour all factor in. This cut makes the model worth testing if you'd previously deprioritised it on cost grounds.

Ezra, Scout AI Team