Diffusion LLM company Inception releases Mercury 3: claimed intelligence leap with no loss of speed
Redwood City, October 19 — Diffusion-based large language model company Inception released Mercury 3, calling it its strongest diffusion LLM to date. According to the company's press release, Mercury 3's total parameter scale is more than 40 times that of Mercury 1, its context window has expanded from 260,000 tokens in Mercury 2.5 to 1 million, and its inference speed remains above 1,000 tokens per second. The company says Mercury 3 surpasses GPT-6 Luna and Claude Haiku 5.5 on multiple benchmarks while being 4 to 8 times faster. The launch price is $0.1 per million input tokens and $0.35 per million output tokens (half the list price). Co-founder Stefano Ermon said the bottleneck in scaling diffusion models lies in compute and data rather than the algorithm itself, and that the company has begun training its next-generation model.