Mercury 3 launches as most capable diffusion LLM, running 4–8x faster

Mercury 3, a new diffusion-based LLM, claims to outperform GPT-6 Luna and Claude Haiku 5.5 on several benchmarks while running 4–8x faster at 1,068 tokens per second. It is over 40x larger than Mercury 1 and supports 1M-token context, priced at $0.20 per million input and $0.70 per million output tokens, with a 50% launch discount. The maker says scaling laws hold for diffusion too, and early partners use it for voice support, tutoring and web navigation.

• Inception claims Mercury 3 diffusion model far outpaces rivals2026-10-21
• NVIDIA launches Kumo Relational foundation model2026-10-21
• Community debates overlooked RTX 3090 performance potential2026-10-21
• Train Microduck to walk in Isaac Lab 3.0 and deploy to real robot2026-10-21
• Cassava Deploys NVIDIA Nemotron on Its CAIMEx Platform2026-10-21