450 TPS on DeepSeek V4 Flash — real user, current rates of $0.12/1M output.

Someone tried our pricing (posted here a bit ago about being 90%+ below DeepSeek’s own new rates) and came back with this: just a real user hitting real throughput on deepseek-v4-flash-0731. We don’t usually lead with numbers like this since performance genuinely varies by load, model, and context , but this one’s worth sharing since it’s unprompted and verifiable. If you want to try it yourself: inferx