Stay ahead on AI.
All in ONE place.

The latest AI news, expert commentary, top tools, and exclusive interviews — curated for you by Matt Wolfe.

Join 250,000+ readers from companies like

GoogleAppleMicrosoftMetaNVIDIAAnthropicOpenAIAmazonAdobeAMDQualcommCiscoa16zSequoia Capital

AI News

The biggest AI headlines, updated throughout the day.

View all news
openai.com·just now

OpenAI GPT-6 Astra Boosts Chip Design for Jalapeño Silicon and AMD GPUs

OpenAI has detailed how its models are accelerating chip design across multiple fronts. GPT-4o helped reduce standard-cell area by roughly 13% in the vector unit and 10% in the matrix unit of OpenAI's custom inference chip, Jalapeño. A collaboration with AMD achieved up to 14% PPA improvement on flagship GPU IP in under five months, with some changes already sent to manufacturing. On RISC-V benchmarks, GPT-6.2 Astra outperformed all prior models. OpenAI is also partnering with Synopsys to train a new GPT-Synopsys model using professional EDA tools.

Tools & Products·17h ago

Perplexity Launches Orchestrator: One API for Multi-Model AI Inference with Auto-Failover

Perplexity has launched Orchestrator, a unified API that connects developers to both open-weight and proprietary AI models with a single integration, unified billing, and automatic same-model failover. It accepts OpenAI and Anthropic SDK requests and routes across providers based on availability, capacity, and latency. Between October 4 and November 2, 2026, Claude and GPT requests through Orchestrator succeeded more often than requests to any single provider. A preview automatic routing feature lets developers assign tasks to specific models using plain-language descriptions.

anthropic.com·18h ago

Claude Fable 5.2 Solves Two Open Black Hole Physics Problems in Quantum Gravity

Claude Fable 5.2, developed by Anthropic, solved two open problems in black hole physics and quantum gravity. In the first, it completed a calculation begun in 2003 that confirmed the Hawking-Page phase transition — the abrupt switch from a particle gas to a black hole — persists even when spacetime loses a definite shape. In the second, it found the exact ceiling for the lightest possible black hole in a simplified two-dimensional universe using the modular bootstrap method. Results were formally verified in Lean in September 2026.

Tools & Products·20h ago

OpenAI Launches ChatGPT for Sales With 30+ Partner Plugins for CRM and Outreach

OpenAI has launched ChatGPT for Sales, a dedicated sales experience featuring a new Sales plugin and over 30 new and expanded partner integrations. The platform connects CRM tools like Salesforce, HubSpot, and Close CRM with prospecting tools such as Apollo and Clay, plus outreach platforms like Outreach and Salesloft. Sales teams can use it to build account lists, prepare call briefs, draft follow-ups, and track pipeline risk. ChatGPT Business and Enterprise admins can enable it via the Admin Console.

Model Release·yesterday

Microsoft Launches MAI Licensed Weights AI Models With Frontier-Level Performance

Microsoft has launched MAI Licensed Weights, allowing organizations to deploy and customize Microsoft AI models within their own Azure environments. Companies can adapt models to proprietary workflows while retaining full control over access, deployment, and upgrade timing. Early results are compelling: AT&T outperformed GPT-6.1 Sol on 5G network tasks at one-fiftieth the cost, Woven by Toyota resolved 44% more code fixes autonomously, and Alaska Airlines improved maintenance report classification accuracy from 81% to 91%.

Model Release·yesterday

Microsoft MAI Image 2.7 Tops GPT Image 2.5 Flare on Benchmarks at Lower Cost

Microsoft has launched MAI Image 2.7 and MAI Image 2.7 Flash, two new image generation and editing models that outperform OpenAI's GPT Image on key benchmarks. MAI Image 2.7 ranks first on Artificial Analysis image editing, beating GPT Image 2.6 Sunburst at one-fifth the cost, and places top two on Arena for both generation and editing. Available in public preview on Microsoft Foundry, MAI Image 2.7 costs about $35 per 1,000 images while Flash costs $10. Microsoft also launched Canvas, an interactive image workspace powered by the Flash model.

Model Release·yesterday

Microsoft Launches MAI Realtime 1 Voice AI at $0.01/Min With 7 New Models

Microsoft has launched MAI Realtime 1, its first speech-to-speech voice AI model, priced at $0.01 per audio minute and targeting enterprise applications. The model outperforms OpenAI's GPT Realtime 2.1 on quality, speed, and price according to public benchmarks and Microsoft's own evaluations. It ranks on the performance-price Pareto frontier for both the Big Bench Audio speech reasoning benchmark and the Tau-Voice agentic task completion benchmark. The launch also includes seven additional new models.

Model Release·yesterday

Microsoft Launches 7 MAI Licensed Weight Models With Frontier Performance at Scale

Microsoft has launched seven MAI licensed-weight models at Microsoft Ignite 2026, targeting frontier performance at lower cost. MAI Code 2 reaches 98% of GPT-6.1 Sol's coding quality at 40% of the price, while MAI Cyber 2 Flash matches GPT-6 Astra at under 2% of the cost. MAI Realtime 1 starts at $0.01 per audio minute, and Decision 1.1 is 50x faster and 79x cheaper than GPT-Sol 6.1. Dragon Copilot Transcribe already powers 10 million patient encounters monthly.

Model Release·yesterday

Microsoft Launches MAI Cyber 2 Flash: Frontier-Level Security AI at Lower Cost

Microsoft has launched MAI Cyber 2 Flash, a specialist AI security model built on the company's proprietary security data and expertise. On the CyberGym-E2E benchmark for real-world vulnerability discovery, MAI Cyber 2 scores 92.3%, matching GPT 6 Astra's 91.5% at less than 2% of the cost per task. It also outperforms GPT 5.6 Cyber on CyberGym-E2E and WebExploitBench at the same reasoning budget. Microsoft plans to integrate the model directly into its security products.

Model Release·yesterday

Microsoft Launches MAI Code 2 & Code 2 Flash: Scalable AI Coding Models

Microsoft has launched MAI Code 2 and MAI Code 2 Flash, two new AI coding models targeting everyday enterprise development tasks. MAI Code 2 delivers 98% of GPT-6.1 Sol's performance at 40% of the cost, and 96% of Claude Opus 5.5's performance at just 5% of the cost, excelling at debugging, multi-file implementation, and visual coding. MAI Code 2 Flash handles high-volume routine tasks like code explanations and rapid iterations, matching GPT-6 Luna at 65% of the cost.

Free subscriber bonus

Get the AI Income Database, free when you join

38+ real ways people are making side-hustle money with AI right now. Each one comes with the playbook and the exact tools to pull it off. No fluff and no course to buy. It's the bonus every new subscriber gets on day one.

  • 38+ proven AI side-hustles, from first dollar to scale
  • The exact tools for each one, picked from the 4,500+ we track
  • Updated as new opportunities appear (subscribers hear first)

Joins the twice-weekly AI briefing read by 250,000+ people. Free forever, unsubscribe anytime.