Inception launches Mercury 3, a diffusion LLM it says tops Haiku 5.5 at 1,000+ tokens per second
First seen on X 9 hours ago@StefanoErmon ♥ 2,9171 views
Inception, a US startup building diffusion-based language models, released Mercury 3.
Per Inception's Oct 19 blog post, Mercury 3 runs at 1,068 tokens per second on NVIDIA GPUs and matches or beats OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5 across a range of benchmarks while running 4 to 8 times faster. Context is 1 million tokens.
Diffusion models refine a whole response in several passes instead of writing one token at a time. Mercury 3 is more than 40 times larger than Mercury 1 but just as fast, the company said. 'Diffusion opens a new path to scaling AI,' co-founder Stefano Ermon wrote on X.
Pricing is $0.20 per million input tokens and $0.70 per million output, half off at launch. It is also available through OpenRouter, Baseten and Vercel. The announcement passed 2,900 likes.
“1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released.”
