🤖 AI HOT Daily · Jun 18, 2026
Grok 4.3 on Bedrock with 1M context, Vercel open-sources Eve AI agent framework, Google's $99 Gemini speaker, Alibaba HappyOyster 1.0, Claude Design integrates with Replit, OpenAI $13B revenue vs $20.9B loss, NVIDIA ENPIRE 8-robot swarm, Anthropic opens Seoul office.
1. Grok 4.3 Launches on Amazon Bedrock
Grok 4.3 is fully available on Bedrock with lowest hallucination rate, 1M token context, delivering 2-10x intelligence per dollar vs other frontier models.
2. Vercel Open-Sources Eve Agent Framework
Apache-2.0 licensed, filesystem-first design where each agent maps to a directory, with persistent execution, sandbox compute, and human-in-the-loop approval.
3. Google Launches $99 Gemini Smart Speaker
Google launches its first Gemini-powered smart speaker at $99.99, with premium AI features via $10/mo Home Premium subscription.
4. Alibaba Cloud Releases HappyOyster 1.0 World Model
Alibaba Cloud launches HappyOyster 1.0 open world model with multimodal input, joint audio-video generation, and real-time user instruction response during generation.
5. Claude Design Update: Brand Consistency & Replit Integration
Claude Design now supports cross-project design systems, syncs with Claude Code, 1M+ users in first week, and integrates with Replit to turn designs into apps.
6. Leaked Docs: OpenAI $13B Revenue, $20.9B Loss
OpenAI's 2025 revenue hit $13.07B but operating loss was $20.92B. ChatGPT has 900M+ weekly active users, ~50M paid. Sora shut down, non-core businesses cut, IPO preparations underway.
7. NVIDIA ENPIRE: 8 AI Agents Control Robots Autonomously
NVIDIA GEAR launches ENPIRE: 8 Codex agents controlling 8 robots running overnight unattended, performing cable tying, fine needle sorting, and GPU installation.
8. Anthropic Opens Seoul Office, Multiple Korean Partnerships
NAVER deploys Claude Code company-wide, Nexon uses Claude Code for game development, LG CNS and Samsung SDS join partnerships.
9. LifeSciBench: AI Life Science Research Benchmark
OpenAI and 173 PhD-level life scientists release a benchmark with 750 real research tasks and 19,020 fine-grained criteria to evaluate models on complex research tasks.
10. Omnigent Open-Sourced: AI Agent Team Meta-Framework
Databricks open-sources Omnigent, enabling an agent team with Claude Code, Codex, Cursor, Pi and more collaborating in a single live session.
Sources and verification
This legacy briefing did not retain its original source URLs. Verify safety, financial, policy, and breaking-news claims with authoritative primary sources before acting on them.