Major AI updates in the last 24h (7 Oct)

OpenAI released Apps SDK letting developers build integrations such as Spotify, Figma and DoorDash directly inside ChatGPT, improving discovery and in-chat experience.
Azure AI Foundry added GPT-image-1-mini, GPT-realtime-mini and GPT-audio-mini models, offering low-cost, high-quality multimodal generation with flexible deployment.
DeepMind introduced CodeMender, an AI agent that has already shipped 72 open-source security fixes and automatically rewrites vulnerable code.

OpenAI's Sora 2 video-generation app hit #1 in the App Store, offering AI-added speech, sound effects and person-insertion for invite-only users.
OpenAI pledged deeper copyright-holder controls for Sora 2, promising takedown mechanisms and potential revenue-share for permitted characters.
OpenAI unveiled Agent Builder, a visual workflow designer for building and deploying autonomous agents without code.

AMD will supply 6 GW of Instinct GPUs to OpenAI, with an option for OpenAI to buy up to 160 M AMD shares (≈10 % stake), a deal projected to yield tens of billions in revenue for AMD.
Deloitte will roll out Anthropic’s Claude chatbot to its 500 k global employees, marking Anthropic’s largest enterprise deployment despite recent AI-hallucination criticism.
OpenAI added in-chat connections to Spotify and Zillow, letting users create playlists or browse homes without leaving ChatGPT.
Microsoft CTO Kevin Scott discussed the company’s AI roadmap at TechCrunch Disrupt, emphasizing partnership opportunities for startups.

Supermemory secured $2.6 M seed round from Cloudflare CTO, Google AI chief Jeff Dean and other OpenAI, Meta and Google execs to build a memory layer that extracts insights from unstructured data for AI applications.

Google dropped the num=100 query parameter that allowed 100 results per page, reducing the amount of data AI models can scrape in a single request.

OpenAI made GPT-5 Pro available via API, targeting high-accuracy use cases in finance, legal and healthcare.
Claude Sonnet models are reported to dominate real-time benchmarks, delivering top-tier reasoning and coding performance.
Google upgraded most Nest hardware with Gemini-powered generative AI, launching $10/month and $20/month subscription tiers for richer notifications.

ChatGPT’s weekly active users grew from 500 M to 800 M, with over 4 M developers integrating OpenAI tools and the platform processing more than 6 B tokens per minute.

A Reddit-sourced study finds that step-by-step “think” prompts can actually degrade performance on simple tasks.
FamilyBench benchmark places Claude Sonnet 4.5 in second place, while Qwen 3-Next exceeds 70 % accuracy on complex tree-relationship tests.

AMD and OpenAI sealed a five-year agreement for 6 GW of GPUs and an option to buy ~10 % of AMD stock, a deal expected to generate tens of billions in revenue for AMD.
Analysts estimate the partnership could bring over $100 B of new revenue for AMD across four years.
NVIDIA introduced the nvCOMP library and a GPU-native Decompression Engine, offloading data-decompression to free compute for LLM training.
Marvell received a Strong-Buy upgrade as Microsoft’s custom Maia AI chip is projected to lift MRVL revenue to $10.5 B in FY26.
IBM and NVIDIA integrated cuDF into the Velox query engine, enabling end-to-end GPU-native analytics for Presto and Spark.

AMD-OpenAI chip deal: 6 GW of GPUs, up to 160 M AMD shares (≈10 % stake), projected tens of billions in revenue.
OpenAI released GPT-5 Pro API, targeting high-accuracy finance, legal and healthcare workloads.
ChatGPT reached 800 M weekly active users, processing >6 B tokens per minute.
OpenAI’s Sora 2 topped the App Store, becoming the most downloaded AI video app.
Supermemory closed a $2.6 M seed round to build a universal AI memory layer.