Inception Mercury 3: Diffusion LLM at 1,068 tok/s
October 19, 2026Inception's Mercury 3 is its most capable diffusion-based language model yet, built for real-time agents, tutoring, and search at over 1,000 tokens per second.
The latest AI news, newest first — updated around the clock by our editors.
Inception's Mercury 3 is its most capable diffusion-based language model yet, built for real-time agents, tutoring, and search at over 1,000 tokens per second.
MiniMax unveiled M3.1 Flash, its fastest agentic AI model yet — and revealed the mystery 'Space Bunny' topping leaderboards was an early version of it.
Serial entrepreneur Bill Gross has launched Idealab.ai, making over 1.2 million business ideas searchable and free, with AI-powered tools to help people build them.
BILL has launched a new AI suite for finance teams, adding invoice risk checks, a centralized AI control hub, chat in beta, and MCP integrations.
Pax8 has launched 8os, an operating system that lets IT partners govern AI agents across small businesses, alongside 8bot, a managed AI coworker for employees.
NVIDIA has released the full Run:ai platform, its Kubernetes GPU orchestration software, as open source under Apache 2.0, so teams can inspect, fork, and self-host GPU scheduling without a paid subscription.
Mistral's new Timestral family offers four zero-shot time-series forecasting models from 5M to 209M parameters that top Google's TimesFM-3 on key benchmarks while running on an ordinary CPU.
Anthropic is placing 5% of its shares, worth about $48 billion, into a new public benefit fund ahead of its planned November IPO, bringing its total equity earmarked for the public good to roughly 20%.
Labelbox has raised more than $100 million in Series E financing to expand its reinforcement learning platform and managed agents for enterprises.
Anthropic has quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI, first reported by Politico.