Mercury
The First Commercial-Scale Diffusion LLM
409 followers
The First Commercial-Scale Diffusion LLM
409 followers
Mercury,from Inception Labs, is the first commercial diffusion LLM. Up to 10x faster than autoregressive models, with comparable or better quality on coding tasks.
This is the 4th launch from Mercury. View more
Mercury 3
Launched this week
Mercury 3 is the most capable diffusion LLM released to date. It outperforms GPT-6 Luna and Claude Haiku 5.5 while running 4–8x faster at 1,000+ tokens per second.
Free Options
Launch Team
Hi everyone!
Mercury 3 is more than 40x the size of Mercury 1, and it still generates at over 1,000 tokens/sec on NVIDIA Blackwell GPUs.
It's a diffusion LLM, generating and refining tokens in parallel instead of predicting them one by one. The context window has also grown from 260K to 1M tokens, with Inception reporting performance around GPT-6 Luna and Claude Haiku 5.5.
Early partners are already using it for browser agents, real-time tutoring, search and customer support. A browser agent might call the model dozens of times for one task, so those seconds can add up quickly.
It's available through an OpenAI-compatible API and OpenRouter.
The browser-agent example is where the speed story gets interesting. I would be curious to see end-to-end task timings with the tool calls included.
1000 tokens/sec is wild 🔥 diffusion for coding feels like the right bet. cheaper than haiku too?
Congrats team! 40x scale with no slowdown on a diffusion LLM is a serious claim, I'd love to see latency side by side with autoregressive models at the same quality. Big fan of the approach, good luck today.