DeepSeek V4 Flash 0731

DeepSeek V4 Flash 0731 just dropped a very loud message into the AI price wars: you may not need to pay luxury prices for top-tier results anymore. The model posted strong scores on ARC-AGI, a test meant to measure how well an AI can solve tricky puzzle-like tasks, while costing just pennies per task. And the crowd response? Equal parts impressed, suspicious, and gleefully chaotic.

The biggest reaction was basically: wait, it does what the expensive models do, but cheaper? One commenter said the results look comparable to a pricier rival and called it “promising,” while another pointed out the almost sitcom-level absurdity that “Max” reasoning somehow costs less than “High” reasoning. Yes, the naming alone had people doing a double take. Another fan favorite reaction summed up the speed of the whole scene: a model that looked exciting a month ago is now being matched for a tiny fraction of the price. In AI time, that’s apparently ancient history.

But not everyone was ready to crown a winner. One skeptic threw cold water on the celebration, arguing that cheap prices can be distorted by investor cash, giant server farms, and behind-the-scenes tricks, so the real comparison may be murkier than the hype suggests. Meanwhile, another commenter stared at the chart and had the most relatable reaction of all: the prices are on a log scale, and it still looks wild. In other words, the benchmark numbers mattered—but the real entertainment was watching the internet argue over whether this was a breakthrough, a bargain-bin miracle, or just another subsidized flex.