DeepSeek V4 applies open-model pricing pressure to closed labs

DeepSeek V4 is the kind of release that makes closed labs suddenly remember to talk about "total cost of ownership" with a straight face. According to Simon Willison's DeepSeek V4 write-up, the preview line includes DeepSeek-V4-Pro and DeepSeek-V4-Flash, both offering a million-token context under an MIT license. The reported prices from DeepSeek's page are incredibly aggressive: $0.14/$0.28 per million input/output tokens for Flash and $1.74/$3.48 for Pro. The open-model ecosystem does not have to win every leaderboard trophy; it just has to be good enough and cheap enough that proprietary pricing starts sounding like a luxury subscription to uncertainty.

When closed labs say "frontier capability," they often mean "please do not ask what this will cost at scale." DeepSeek’s move makes cost part of the positioning. A bigger Pro model and a cheaper Flash model give teams routing options to spend more on hard analysis and less on routine extraction or coding helper tasks. Long context can hide failures beautifully, so the model may read the whole haystack and still hand you the wrong needle, but open models offer costs you can inspect and negotiate with. Every strong open model compresses the premium tier from below by making buyers ask why a routine workload is paying flagship prices when a controllable model is sufficient. If Flash handles a large share of routine work at the cited price, the bigger story is routing economics, which is where open models keep punching above their keynote weight.