DeepSeek cuts Flash prices up to 57%, folds vision into base model, retires V4 Pro
By Rob Litterst, Co-founder, PricingSaaS
Published
DeepSeek’s pricing page · Change detected
What changed
DeepSeek cut Flash-tier token prices across the board: cache-hit input down 57%, cache-miss input down 32%, output down 9%. The separate flash-vision-exp model was retired and merged into the base Flash model, which now natively supports vision. Pro-tier pricing is unchanged. Separately, DeepSeek will retire the V4 Pro model starting September 14, 2026 at 12:00 Beijing time, rerouting all deepseek-v4-pro requests to V4.1.
Why it matters
Folding vision into the base Flash model while cutting its price suggests DeepSeek is consolidating its lineup ahead of the V4 Pro retirement rather than just chasing usage. Watch whether V4.1, the model absorbing rerouted V4 Pro traffic, gets its own price move once the September 14 cutover completes.
In context
DeepSeek's pricing has been unusually volatile for a model this young in the index: it announced 2x peak-hour surge pricing in early August, walked that back a week later with a vague warning of a broader increase, then actually restructured to peak/off-peak billing that roughly doubled Flash rates, before carving out weekend exemptions and adding a vision model later that month. Today's move reverses direction entirely, cutting Flash prices instead of raising them, while merging the vision variant it had just added into the base model. Credits and usage-based billing changes are common across the index, 583 in the last six months across 324 companies, led by HubSpot (27) and Datadog (12), but few companies have cycled through increase, restructuring, and cut this fast.
What people are saying
“DeepSeek released their smallest model today. It was so good they're retiring their flagship. V4.1 Flash is the first model in their new architecture family, and it beats V4 Pro on performance, cost and speed. From Monday, every call to their pro model gets routed to the flash model and billed at flash prices. They retired their flagship because the cheap one beat it. It's also DeepSeek's first model with native vision. Text-only DeepSeek is over. 1M context, and they cut API prices again on top of it. For anyone building products on these APIs, this is the part to watch. The model that wins isn't the smartest one. It's the one fast and cheap enough to live inside a real product. Speed and price beat benchmarks, every time. Give me a slightly worse model that responds in half a second over a genius that makes my users wait. Every single time.”
View original“the biggest news in this launch is not on the chart. deepseek is retiring V4-Pro, its flagship, next week. from sept 14 every request for the big model gets routed to this small one, at the small one's price. the small model beat the big one 😆”
View original“DeepSeek is retiring V4 Pro into V4.1 Flash in four days... flash is the new default.. as Pro gets redirected Same API name, cheaper bill - new architecture, native vision, 1M context, open weights.... they also cut KV cache to 1/4 of the last Flash so agents don’t eat HBM alive on their own benches Flash is ahead of Pro on DeepSWE, cyber and automation still...killing your own Pro because Flash was better is a rare move only we can see from Chinese labs”
View original