DeepSeek Drops V4 Flash Vision Exp — 1M Context, Image Support, $0.44/M Tokens
What shipped? DeepSeek released an experimental vision-enabled variant of DeepSeek V4 Flash, now available on OpenRouter as deepseek/deepseek-v4-flash-vision-exp.
What changed? The new model adds image understanding to the V4 Flash architecture while retaining its text capabilities including agents, reasoning, and world knowledge. It runs a 1,048,576 token context window with up to 384,000 output tokens. Pricing lands at $0.44 per million input tokens and $1.32 per million output tokens — the same competitive rate as the base V4 Flash 0731. Created August 21, 2026 (timestamp 1787311563). A dated snapshot is also available at the -20260821 tag for reproducible results.
Why does a builder care? If you've been running V4 Flash for text-only agent workloads and wanted to add vision capability without changing architectures or paying a premium — this is the same price point with multimodal support. 1M context means you can dump entire codebases, documentation sets, or image corpora into a single call. The "experimental" tag means it ships fast and may see rapid iteration. Anyone building vision-enabled agents on a budget should test against this immediately.