Labs

Inception Labs has introduced Mercury 3, which it claims is the most capable diffusion-based LLM released to date. The company says the model outperforms GPT-6 Luna and Claude Haiku 5.5 on a range of benchmarks while running 4-8x faster at speeds exceeding 1,000 tokens per second.

Models • Hacker News

Dream Labs • Seed

Dream Labs, which offers an API for its Dream-1 robot foundation model trained on human videos, raised a $90 million seed round. The round was led by B Capital and Kindred Ventures.

Microsoft is advising enterprises to closely monitor AI agents used by employees, equating them with insider threats. The company argues that frontier models cannot be trusted when operating on systems not controlled by AI labs like OpenAI and Anthropic.

AI Safety • TechCrunch AI

Hugging Face's policy chief expressed skepticism about 'embedded evaluators,' the concept of placing supposedly neutral third-party observers inside AI organizations. The idea has gained popularity as AI labs face safety incidents and concerns over potential harms.

AI Safety • TechCrunch AI

ElevenLabs released two new speech-to-speech models designed for real-time AI conversations. The v1 Max variant reportedly scores 98.8 percent on the Big Bench Audio benchmark with a 0.57-second response time, which ElevenLabs says no faster model on the Artificial Analysis leaderboard has matched.

Models • The Decoder

AI labs are increasingly buying data from defunct startups to train their models, turning the digital archives of failed companies into a valuable commodity. One founder, Keelan Miskell, netted roughly $25,000 for his crypto startup's code base and data.

AI Industry • MIT Technology Review