AI Pulse · Business Models & Winners

DeepSeek’s new pricing kicked in today, and it’s a real jump. Output tokens on its V4 Pro model went from $0.87 per million to $3.96 per million during peak hours. Off-peak, you’re looking at $1.98 per million. That’s roughly four to five times what people were paying just last week, depending on when you run your workloads, according to the original report.

If you built anything on the assumption that dirt cheap AI was the new normal, this is your reminder that it wasn’t. DeepSeek’s rock bottom prices were never a stable floor. They were a land grab, priced to pull developers and companies away from OpenAI and the other big labs while DeepSeek built market share. Now that it has attention and usage, the bill is catching up to something closer to real cost. Expect other providers to test similar moves once they feel secure enough in their user base to stop subsidizing you.

For an owner, the lesson isn’t about DeepSeek specifically. It’s about how you budget for AI at all. If a chunk of your operations, customer support, content, coding, research, runs through a single cheap model, you’re exposed to whoever set that price staying generous. Peak and off-peak tiers also mean your costs now depend on when your team or your customers actually use the tool, not just how much.

The practical move is to treat AI spend the way you’d treat a cloud hosting bill. Track usage by time and by task, know which models you can swap out if pricing shifts again, and build in some flexibility rather than locking your whole workflow to one vendor’s rate card. This kind of usage tracking and model switching is exactly the kind of thing we build into an AI command centre, so cost shifts like this don’t catch you off guard.