June 8, 2026
Too Fast Too Curious
MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second
Xiaomi says its giant AI is lightning-fast, and the comments instantly lost their minds
TLDR: Xiaomi says its massive new AI can answer at blistering speed, a big deal because faster replies could make AI feel instant instead of annoying. Commenters were excited, skeptical about the economics, and immediately tested whether the model could be trusted on sensitive topics.
Xiaomi just rolled in shouting that its new MiMo-V2.5-Pro-UltraSpeed can spit out text at a jaw-dropping 1000 tokens per second — basically, an AI that types so fast the company is pitching it as less like a tool and more like a brain sidekick. The catch? It costs 3 times more, access is application-only, and the free trial is locked to a tight two-week window. So naturally, the community reaction was a mix of hype, side-eye, and chaos.
Some commenters were fully in their "shut up and take my money" era. One user simply posted "boom!", which honestly captures the vibe better than half the press release. Others said this is exactly where AI should be heading: not necessarily making models “smarter,” but making them fast enough that humans don’t lose their train of thought waiting around. That sparked a mini-consensus that speed might be the real unlock for coding helpers and digital assistants.
But the thread wasn’t all cheering. One commenter immediately brought up a political censorship test, checking whether the model answers a sensitive historical question correctly — and said, surprisingly, that it passed. That gave the launch a little extra drama: it wasn’t just about speed anymore, but trust. Meanwhile, another user did the classic forum accountant routine, arguing the pricing seems almost suspiciously cheap, hinting Xiaomi’s profit margins might be getting squeezed. In other words: the launch dropped, and the comments turned it into a full-on mix of speed worship, skepticism, and meme-worthy disbelief.
Key Points
- •Xiaomi announced MiMo-V2.5-Pro-UltraSpeed, a 1-trillion-parameter model it says exceeds 1000 tokens per second decode speed.
- •The release was presented as a collaboration with TileRT.
- •The UltraSpeed API is priced at 3× MiMo-V2.5-Pro and is described as delivering about 10× the generation speed.
- •Access is application-based and limited to a trial window from June 9 to June 23, 2026, due to constrained high-speed inference resources.
- •Xiaomi highlights use cases including parallel reasoning methods, coding agents, and time-sensitive applications such as anti-fraud, bidding, dialogue, and medical assistance.