Friday, Nov 6, 2026
AI;DR
2026-11-06 · 11 stories- openaiOpenAI launched GPT-6 Cyber, saying it beats GPT-6 Astra on cybersecurity benchmarks and finds vulnerabilities more effectively. The model is classified “Critical” but is reportedly less likely to go rogue.
- anthropicAnthropic’s planned IPO faces valuation pressure and possible delay as investors retreat from the earlier $1.5T–$2T expectations. Slower growth, higher infrastructure costs and a new operating loss weakened the case.
- anthropicMETR says Anthropic’s models are improving about twice as fast as during March 2024–October 2025. Internal AI use reportedly drives roughly 80% of that acceleration, showing how tools can compound research productivity.
- regulationAn election-linked AI selloff erased trillions from technology and crypto markets: Nasdaq fell about 5%, Japan’s Nikkei entered a bear market and crypto lost $150B. Fears of tighter AI rules and slower data-center growth drove the rout.+4
- googleGoogle released Gemini Omni 1.2 Flash with native 15-second video generation, stronger consistency and conversational editing. Its 30% price cut to $0.07 per second aims to make longer AI clips more affordable.+1
- metaMark Cuban alleges Facebook knowingly ran AI deepfake scam advertisements, while documents reportedly put related fraud revenue at $16B. The claims intensify scrutiny of platform safeguards and ad-review accountability.
- llmThinking Machines Lab released Inkling 2 Small, a 594B-parameter mixture-of-experts model with 17B active parameters. It scores 31 on the Artificial Analysis Intelligence Index, with Apache 2.0 weights planned.
- Baidu opened weights for ERNIE-5.5-Flash-Base, a 280B-parameter MoE with a 1M-token context window. Its 82.3% HumanEval+ score beats DeepSeek-V4-Flash-Base, while vLLM and SGLang support broadens deployment.
- llmTencent expanded Hy4 inference capacity tenfold after heavy WorkBuddy demand and extended free Hy3 access. Hy4 leads China-based labs on agentic coding, but its 89% hallucination rate highlights serious reliability limits.
- nvidiaNVIDIA introduced three open Cosmos 3 models for interactive world simulation. The release gives developers new building blocks for AI-generated environments and embodied applications, strengthening its open-model ecosystem.
- agentPerplexity launched its own AI agents for workplace tasks, entering a market led by OpenAI and Anthropic. The move broadens Perplexity beyond search and raises competition around practical, task-oriented assistants.