DeepSeek Releases V4.1-Flash: 70% Output Price Cut, Native Vision
On September 10, 2026, DeepSeek launched DeepSeek-V4.1-Flash (deepseek-flash), with pricing effective 04:00 UTC.
Three Key Changes
- Cheaper: ~70% output price cut vs the prior flagship tier; off-peak half-price windows; cache-hit input at $0.003/1M tokens
- Sees images: native multimodal vision input, a first for the Flash line
- Auto-migration: from September 14,
v4-procalls route to V4.1-Flash
The 552B-parameter MoE approached or beat V4 Pro (0813) on several benchmarks, with reported ~88% reduction in long-context cache disk usage.