We gotta give the DeepSeek team a lot of thanks for shaking up the AI market and industry. Each release pushes performance and efficiency to the max and scares the bejesus out of their competitors. We need more companies and researchers like them

Late to this thread, but yeah totally agree

The “fraction of the price” claim holds up on the current list cards, with one caveat: V4 is peak/off-peak, not a flat sticker.

At peak, V4 Pro is $1.32/M in and $3.96/M out (cache hits $0.044). Off-peak (every hour outside 01:00-04:00 and 06:00-10:00 UTC on weekdays, plus all weekend) those are halved. V4.1 Flash is even lower: $0.30/$1.20 peak and $0.15/$0.60 off-peak, with cache hits at $0.006. For comparison, frontier OpenAI/Anthropic tiers are still in the $2-$10 in / $10-$50 out range, so Flash especially is roughly an order of magnitude cheaper per token.

Where people get surprised: (1) output is still 3x input on Pro, so chatty agents aren’t “almost free”; (2) if your traffic lands on-peak, you pay the full card; (3) cache only helps if the prefix is byte-stable. I keep a hand-checked DeepSeek sheet with the peak and off-peak rows side by side at DeepSeek API Pricing — Latest Models & Weights · AI//COST (disclosure: I work on that site).