AI Morning Brief — 9 July 2026
SpaceXAI’s release of Grok 4.5 dominated overnight coverage, drawing sustained commentary on benchmarks and pricing, while OpenAI completed the GPT-Live voice model rollout to all paying subscribers and Anthropic published safety research on dual-use AI capabilities.
Top stories
- Grok 4.5 ships. SpaceXAI released Grok 4.5, a model trained on NVIDIA GB300 NVL72 hardware and positioned for coding and agentic work. It prices at $2/$6 per million input/output tokens, ranks fourth on the Artificial Analysis Intelligence Index (behind Fable 5, GPT-5.5, and Opus 4.8), and places first on Harvey’s Legal Agent Benchmark. A 1M-token context window is expected within a week. via MarkTechPost
- GPT-Live fully rolled out. OpenAI announced that GPT-Live, its next-generation voice model, is now available to all Go, Plus, and Pro ChatGPT subscribers, with free-tier access in progress. The model delegates harder tasks — web search, extended reasoning — to GPT-5.5. via Simon Willison
- Anthropic publishes dual-use safety research. Anthropic and AE Studio released work on managing AI capabilities that are both beneficial and dangerous, exploring an off-switch framework for capabilities such as virology knowledge that can serve medical research or enable harm. via Anthropic
- Claude apps gateway on AWS. AWS announced a self-hosted Claude apps gateway: a control plane giving organisations centralised access, cost, and policy management for Claude Code and Claude Desktop via Amazon Bedrock. via AWS Machine Learning Blog
- Lovable eyes $13.2B valuation. The AI app-builder is reportedly in talks to raise $300M led by Menlo Ventures, which would double its current valuation. via TechCrunch
- Prime Intellect raises $130M. The enterprise agent-training infrastructure startup closed a Series A to let organisations build and run agentic AI systems without relying on frontier labs. via TechCrunch
- Grok CSAM lawsuit. A lawsuit claimed a Grok user produced approximately 7,000 child sexual abuse images; xAI is alleged to have reported only one prompt to authorities. The case adds to scrutiny over content moderation at xAI. via Ars Technica
Who shipped
SpaceXAI drove the cycle with Grok 4.5, backed by heavy promotion from Elon Musk and rapid third-party benchmark comparisons placing it in the upper tier for coding tasks. OpenAI completed the GPT-Live rollout and posted an analysis questioning the reliability of SWE-Bench Pro as a coding evaluation. Anthropic published dual-use safety research and launched a new AWS enterprise integration. Pika released Director’s Suite, an agent-based filmmaking tool powered by Claude that generates images, video, and music within a single editing timeline.
Open-source pulse
Ant Group’s Robbyant released LingBot-VLA 2.0, an Apache-licensed 6B vision-language-action model pretrained on roughly 60,000 hours of robot trajectory and egocentric video data. Chinese startup MiniMax announced plans to open-source a 2.7 trillion-parameter model later this year, which would be among the largest publicly released weights to date. The llama.cpp project published builds b9916 through b9934 overnight, covering OpenCL MoE optimisations, WebGPU attention tuning, and new streaming ASR model support.
Money, infra & hardware
Lovable is reportedly in talks to raise $300M led by Menlo Ventures at a $13.2B valuation. Prime Intellect closed a $130M Series A for enterprise agent infrastructure. Google DeepMind added background execution and MCP support to Gemini API managed agents. GitHub Copilot received enterprise MDM configuration delivery for VS Code and CLI.
Quiet corners
DeepSeek was absent from primary sources, appearing only in speculative posts about a forthcoming V4 general-availability release. Meta and the broader Llama ecosystem were quiet throughout the window.
By the numbers
- 360 stories in 24 h across 30+ sources
- Most-mentioned model: Grok
- Most-mentioned lab: xAI / SpaceXAI
- Notable absences: DeepSeek, Meta
Compiled by AI Feed’s editor from all 360 headlines published on 9 July 2026.