Inception releases Mercury 3, a diffusion LLM it says beats GPT-6 Luna and Haiku 5.5, at $0.10 / $0.35 for launch

Inception released Mercury 3, its largest diffusion language model, on October 19. Co-founder Stefano Ermon announced it on X at 16:42 UTC as "the most intelligent diffusion LLM ever released," Inception's blog post "Diffusion at Scale: Introducing Mercury 3" went up the same day, and OpenRouter already lists it as inception/mercury-3.

THE PRICE

$0.10 per million input tokens and $0.35 per million output, which Inception calls a 50% launch discount, on its homepage and in Ermon's thread. OpenRouter and Vercel's AI Gateway list the same price, with cached input at $0.01. Inception has not said when the discount ends; one other gateway, Kilo, already lists double, $0.20 / $0.70.

THE MODEL

A diffusion LLM, which drafts many tokens in parallel and refines them rather than writing one token at a time. The catalogs list a 1M-token context window and up to 131,072 output tokens. Ermon says it is more than 40 times the size of Mercury 1 and still runs at over 1,000 tokens per second.

WHAT INCEPTION CLAIMS

That Mercury 3 outperforms OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5 on quality. Those are Inception's own claims; no independent board has scored it yet. Ermon names early users in code review, education and browser use.

WHERE TO GET IT

The Inception API Platform, OpenRouter, Baseten and Vercel.

WHAT WE DO NOT KNOW

The full price after the launch discount, and how Artificial Analysis and the other boards will score it. We had no watch on Mercury 3 before today, so it reached our board late; it is listed now, and we are watching for the next Mercury.