Company@X — 2026-07-24#

Signal of the Day#

A massive coalition of tech heavyweights—including Microsoft, NVIDIA, Meta, a16z, and Hugging Face—published a joint letter defending open-weight AI models. This coordinated policy push frames open models as an essential requirement for global competitiveness, cybersecurity, and sovereign innovation, aiming to steer the regulatory narrative away from locked-down, closed-ecosystem mandates.

Key Announcements#

Anthropic · Source Anthropic introduced Claude Opus 5, describing it as a thoughtful and proactive model. It reportedly approaches the frontier intelligence of competitor Fable 5 but operates at half the price, signaling a fierce pricing war at the most advanced model tier.

xAI (Grok) · Source Grok has officially integrated into Google Workspace via a unified add-on for Docs, Sheets, and Slides to automate drafting and reporting. Concurrently, xAI highlighted that Grok 4.5 outperformed Gemini Flash 3.6, GPT 5.6 Terra, and Sonnet 5 in a 150,000-invoice zero-click processing benchmark run by Ramp.

AMD · Source AMD’s new MI455X GPU ships with 432GB of HBM4, packing more memory than five NVIDIA H100s combined. With early access, Hugging Face reports that a standard 4-GPU server (1.7TB memory) can now comfortably hold the FP8 weights of a trillion-parameter model in a single machine.

Induction Labs · Source The YC-backed startup emerged from stealth to launch “imagination models,” a new foundation architecture that learns from internet-scale, unlabeled video. Their first model, Photon-1, learned to use a computer entirely by watching 18 years’ worth of screen recordings.

OpenCode · Source The open-source AI coding agent revealed explosive scale, hitting 4.6 million weekly active users, processing 7 trillion tokens daily, and reaching roughly $40M in annualized revenue. The platform’s enterprise growth was reportedly accelerated by an Anthropic API clampdown that drove massive developer demand for model-agnostic tooling.

Google · Source Google published gmn, a parametric differentiable 3D head model that operates efficiently on a CPU. The model allows for every facial expression and movement to be differentiable and malleable, with an interactive demo currently hosted on Hugging Face Spaces.

Also Noted#

• AWS (Source): AWS launched a new Local Zone in Athens, Greece, offering single-digit millisecond latency for localized S3 and EBS workloads.
• Nebula Security (Source): The company launched VEGA, a YC-backed AI cybersecurity agent designed to review source code and proactively investigate supply-chain attack paths before products ship.
• Heart Aerospace (Source): The company’s electric X1 aircraft has officially been cleared for flight.
• EliseAI (Source): Real estate market data shows the median AI voice call with renters has increased to 1 minute and 40 seconds, matching human call lengths and effectively closing the conversational “uncanny valley”.