DeepSeek V4.1 Flash shipped Sep 10. Four days later it's already #6 on OpenRouter by real token volume, ~6T tokens/week. Not a benchmark, actual usage.
Here's the part that matters: it's one of the pricier open-weight models there at 0.15/0.60 per M. So it didn't climb by being free. People are paying to run it.
ngl this is the "cheap backend that actually works" thesis getting field-tested. 1M context, open weights, and devs routing real traffic to it within a week of launch. The "China model = cheap but meh" line keeps aging badly.
Team "rank by benchmark" or team "rank by what people actually run" — which board you trust?
Top comments (0)