GPT-5.4 vs DeepSeek V4 Pro: A Photo Finish Up Top
GPT-5.4 and DeepSeek V4 Pro are effectively tied at the top of LMArena's Text leaderboard — 1,452.22 to 1,451.00, with overlapping confidence intervals. If you're picking on human preference alone, that's a statistical dead heat. The real difference is who you are and what you're optimizing for, so let's break it down.
The scoreboard
The October 2 LMArena Text snapshot puts GPT-5.4 at 1,452.22 and DeepSeek V4 Pro at 1,451.00. Those numbers are close enough that the reported confidence intervals overlap, which means the ordering isn't stable. On any given day, a fresh batch of votes could flip the ranking. Calling one the "winner" would be overreading the data.
What that tie really means
A one-point gap on a human-preference leaderboard isn't a performance gap you'll feel in real use. GPT-5.4 might handle certain prompts with a bit more polish; DeepSeek V4 Pro might win on others. The overlap in confidence intervals says these differences are mostly noise, not signal. If you're choosing between them, you need to test on your own tasks rather than trust a single ranking.
The missing piece: price
Neither model's official current pricing was included in the benchmark data we have, so we're not going to invent numbers. That's a real blind spot. If one of these comes in significantly cheaper per token when pricing is announced, that could break the tie faster than any leaderboard point.
Verdict: which one should you pick?
Pick GPT-5.4 if you want the exact top slot on the leaderboard and don't mind paying for a name-brand frontline model. Pick DeepSeek V4 Pro if you want essentially the same human-preference results and you're open to a challenger that may offer a better deal. On the evidence available, this one is too close to call on quality alone.