DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis

DeepSeek’s new V4 Flash 0731 arrived with a very simple flex: it looks smart, talks a lot, and costs less than many rivals. The benchmark page says it scores 50 on the site’s intelligence ranking, well above the typical model in its price bracket, while charging just $0.14 for input and $0.28 for output per million tokens. In plain English: people are reading this as a model that gives you a lot of brainpower without setting your wallet on fire. And yes, the gigantic 1 million token context window means it can handle absurdly long prompts, which only adds to the “budget beast” vibe.

But the real show was in the comments, where the launch immediately turned into a mix of price-war chest-thumping, benchmark paranoia, and future-hype fever. One commenter cut straight to the drama, claiming DeepSeek has “already beat Luna on price/task, by about 2x”—basically the AI version of yelling “check the scoreboard.” Another mini-scandal popped up when users tripped over a bad link, producing the most internet moment possible: before debating the model, they had to debate which URL actually worked. Then came the nerd fight over whether benchmark numbers match real-world behavior, with one commenter openly questioning if the token-count data is wrong because another model, Kimi K3, feels much more long-winded in practice. And looming over all of it was the classic fan-fiction energy of AI comment sections: if Flash is this good, is a bigger, stronger Pro model about to drop and start swinging at Opus? The mood is equal parts impressed, suspicious, and wildly ready for the next round.