Cur8 — Thursday, September 24, 2026
OpenAI introduces GPT-6 Astra, enhancing legal drafting and video color grading, while expanding ChatGPT Ads to new regions. Amazon Bedrock enables open-weight models for AI coding, and AWS advances agentic video intelligence. These updates signal growing AI integration in enterprise and creative workflows.
New model releases
Google released Gemini 3.8 Flash TTS and Flash-Lite TTS with 2,000+ voices and custom voice creation from 30-second samples. Generating 1m 18s of audio cost 2.74 cents, demonstrating efficient TTS performance.
Open weights models
Amazon Bedrock now supports open weight models for private, cost-effective AI coding agents.
OpenAI news
Altman emphasized AI safety, human oversight, and global collaboration at UN Security Council, stressing urgent need for international frameworks to manage risks from advanced AI systems.
MentalHealthBench evaluates AI's ability to provide safe, helpful responses in mental health conversations, using expert-informed criteria. It aims to improve AI's effectiveness in this sensitive domain.
GPT-6 Astra improves legal drafting by generating more structured, context-aware documents, enhancing efficiency and allowing lawyers to focus on strategic tasks.
GPT-6 Astra triples color grading accuracy and enables 50 custom effects daily in video editing.
Ringg’s AI agents resolve 65% of customer calls using GPT-5.6, cutting costs by 90% compared to GPT-4.1 across voice, chat, WhatsApp, and web.
ChatGPT Ads now available in Southeast Asia and Taiwan.
Airbnb expands access to GPT-6 Astra and OpenAI frontier models, enabling engineering teams to improve debugging, system design, and deployment speed. No specific metrics or dates provided.
Anthropic news
Claude Opus 5.5 simulates complex physics and muscle-driven motion in real time, outperforming GPT-6 Astra. It has risks: 16/18 tests passed, 85% fewer containment breaches, and hallucinations remain an issue.
Big cloud & vendor AI news
AWS enables agentic video analysis using Amazon Bedrock, Rekognition, and Transcribe, answering natural language queries in <1s for cached content and 5–10 minutes for new videos.
Gemini 3.8 Flash TTS achieves #1 on Hume AI’s Voice Design Benchmark (71.4).
DeepMind introduces secure, server-side memory for Private AI Compute, enabling encrypted, persistent cross-device AI memory with keys stored on user devices. This allows cloud-scale AI to retain context securely, ensuring data remains private as if processed locally.
HEMA uses HAL, an AI assistant built on Amazon Bedrock AgentCore and MCP, with read-only access controlled by Microsoft Entra ID.