Running Your Business
A Different Way to Make AI Faster: Mercury 3 Explained

On October 19, a company called Inception launched Mercury 3, an AI that works differently from most.
How it is different
Most AIs write one word at a time. Mercury writes many at once, then fixes them. This is called a diffusion language model.
Speed
Inception says it makes about 1,068 tokens (small chunks of text) per second on NVIDIA graphics chips. It says that is 4 to 8 times faster than speed-focused models like GPT-6 Luna and Claude Haiku 5.5.
What it costs
- $0.20 to read and $0.70 to write, per million tokens
- A 50% launch discount, which brings that to $0.10 and $0.35
- 90% off for repeated input
- 100 million free tokens for new accounts
You can find it on the Inception API, OpenRouter, Baseten, Vercel and Mercury Chat.
What we could not confirm
Inception says Mercury 3 matches or beats rivals on a range of tests, but it gave no scores. It shows one live test, 50% completion on 50 browser tasks. The model's inner files are not published.
What this means for you
If your AI feels slow, ask whether a faster model could do the same job. Test it on a real task first.
Sources
Want a website your customers can trust, kept safe and up to date? Get a free quote.
Questions about this topic
Is Mercury 3 better than other AI models?
Inception says so on a range of tests but gave no scores. Test it on your own work.
Is there a free trial?
New API accounts get 100 million free tokens, according to Inception.


