AI Pulse · AI Trends Pulse

DeepSeek just released V4 Flash 0731, and it beats its own previous model across all nine benchmarks while costing less than fifteen cents per million input tokens. The architecture is a 284 billion parameter mixture of experts model that activates only 13 billion parameters per query, handles a million token context window, and charges $0.14 in, $0.28 out.

This matters because the cost curve for capable AI keeps dropping. A year ago, models with this kind of performance would have cost five times as much to run. Now you can process a million tokens, roughly 750,000 words, for about fourteen cents. That changes the math on what you can afford to automate. Document analysis, customer support routing, code review, anything that used to be too expensive to throw AI at becomes viable.

The model topped Hacker News with over 700 upvotes and landed at number two on Hugging Face’s trending list this week, according to the model card. Developer attention signals where practical tooling is heading. When technical communities pile onto a release like this, it usually means the performance claims hold up under real use.

For operators, the takeaway is simple. If you wrote off an AI use case six months ago because the token costs didn’t pencil out, run the numbers again. The mixture of experts architecture means you get large model capability at small model prices. That gap between what AI can do and what it costs to do it keeps shrinking, and the projects that didn’t make sense in Q1 might be worth a second look now. This is exactly the kind of shift we track and build into tools like the Omni Command Centre, so teams can swap models as economics improve without rewriting their workflows.