DeepSeek is adding time-of-day pricing to its upcoming official V4 release, leaving baseline rates untouched while doubling charges only during Beijing's designated peak windows on both Flash and Pro tiers.
DeepSeek V4 is coming. Mid-July. Yeah yeah you might think we had V4 for over 2 months already, but no, that was "preview of V4". They expect a lot of demand, and introduce a new mechanism: peak hour pricing. It doubles what we have now. The baseline is the same though.
v4 正式版价格翻倍性能提升应该不小吧
Off-peak costs for cache hits, misses, and outputs remain exactly where they are now for V4-Flash and V4-Pro, with the doubling limited strictly to the 9-12 and 14-18 Beijing slots.
The 1M token context and 384K max output stay standard across both tiers along with JSON output and tool calls, though the exact final date inside mid-July is still unannounced.
Users are excited for DeepSeek's official V4 launch with peak pricing due to advanced model access and regional non-peak benefits, while some criticize the pricing shifts as ruthless.
No Digg Deeper questions have been answered for this story yet.
The phrasing suggests there are no new features, just continued infra optimization and model polish. Oh well
DeepSeek V4 is coming. Mid-July. Yeah yeah you might think we had V4 for over 2 months already, but no, that was "preview of V4". They expect a lot of demand, and introduce a new mechanism: peak hour pricing. It doubles what we have now. The baseline is the same though.
get ready for cheap Opus level intelligence
DeepSeek V4 is coming. Mid-July. Yeah yeah you might think we had V4 for over 2 months already, but no, that was "preview of V4". They expect a lot of demand, and introduce a new mechanism: peak hour pricing. It doubles what we have now. The baseline is the same though.
Previously they had: higher pricing for R1 vs V3; off-peak discounts; "permanent discounts" and a number of flat rate cuts. Off-peak discount is structurally identical to this ofc. in any case I approve, they need revenue. Current pricing:
DeepSeek V4 is coming. Mid-July. Yeah yeah you might think we had V4 for over 2 months already, but no, that was "preview of V4". They expect a lot of demand, and introduce a new mechanism: peak hour pricing. It doubles what we have now. The baseline is the same though.
@teortaxesTex Completely forgot Pro & Flash were technically previews, hopefully remaining at the same size
@teortaxesTex Here is the thing. During the day in China it's a peak hour. During the day in 'Murica it's also peak hour. So it leaves not so many hours where it is cheap.
@teortaxesTex Good thing i spent the whole weekend doing a bunch of token optimizations
@teortaxesTex Can we not expect any performance improvement?
@teortaxesTex 🤦
No. it'll be worse because Dario has banned their distillation accounts. Also and more expensive because no EUV in China. Whale is beached, sadly… They'll need a few months until their compliance forms for Mistral accounts are processed to begin distilling from Le Chaton Fat
@teortaxesTex Can we not expect any performance improvement?
@teortaxesTex I got the email as well, hope it will be more token efficient lol
@teortaxesTex So, I see, it's not Deepseek v4.1, but just Deepseek v4 that is being released in July.
@teortaxesTex DeepSeek给海外用户也同样享受北京时间计算的高峰期？总有时区是完美享受到工作时间但是原价的Token吧
@teortaxesTex 按照这个速度可能最多只有 v4.2，然后明年就是全新的 v5 版本
@teortaxesTex So that's why I was sent a Chinese-language email from DeepSeek just now
@scaling01 the real shift is the floor
entirely new tier of capability becomes accessible to everyone overnight
@teortaxesTex 涨价预警先行这波操作有点狠
@teortaxesTex Excited for the official V4! Any chance of bigger context windows or new reasoning modes in the mid-July release?
@teortaxesTex Peak-hour pricing is an interesting tell. If baseline stays the same, the big question is whether users can actually shift workloads enough to avoid the premium.
@teortaxesTex Seems more like an effort to balance load and push traffic to different times. The most important price, cache hit is still dirt cheap.
@scaling01 I am ready 🫡🚀