Live Feed
US Nears Voluntary AI Standards Deal: 30-Day Pre-Release Review, Classified Benchmarks, Meta Not at the Table
The Trump administration is in advanced talks with OpenAI, Anthropic, Google, Amazon, and Microsoft to formalise a voluntary framework for frontier model releases. The deal would give government a 30-day review window before public launch. Meta has not joined current negotiations.
Fable 5 Returns Worldwide as GPT-5.6 Sol Stays Government-Only: A Benchmark Split With an Access Wall
Claude Fable 5 went back on sale globally July 1. GPT-5.6 Sol is still limited to roughly 20 government-vetted organisations. The two leading models now hold different benchmarks and sit behind different access gates.
Meta Launches AI Cloud to Sell Excess Compute — META Surges 10%, CoreWeave Falls 10.8%
Meta is building a cloud business to monetise its idle AI infrastructure, letting developers rent Llama-hosted model access and raw GPU capacity from Meta's own data centres. The announcement sent Meta shares up over 10% and cratered AI cloud competitors: CoreWeave dropped 10.8%, Nebius fell 12.4%.
Nvidia Puts Its Balance Sheet Behind AI Clouds: 210,000 GPUs on Revenue Share With Sharon AI and Firmus
Nvidia is fronting hardware to AI cloud operators in exchange for a cut of token revenue, a structural departure from its chip-sales model. First partners Sharon AI and Firmus will deploy 40,000 GB300s and 170,000 GPUs respectively, anchored by a 360-megawatt Firmus campus in Batam, Indonesia.
OpenAI Proposes 5% US Government Stake Worth $42.6B — Political Insurance at $852B Valuation
Sam Altman has floated giving the Trump administration a 5% ownership slice of OpenAI, reportedly worth $42.6B at the company's current valuation. The move is framed as a way to share AI upside with the public while blunting regulatory and political pressure ahead of its IPO.
xAI Launches Voice Agent Builder in Beta: No-Code Call Centre on Grok Voice at $0.05/Min
xAI's new no-code platform lets anyone build a production voice agent using Grok Voice Think Fast 1.0 in under two minutes. 80+ preset voices, voice cloning from 2 minutes of audio, and integrations with Notion and Gmail. API billed at $0.05/minute.
Z.ai Ships ZCode: GLM-5.2 Coding Agent Topped HN at #1, Runs From WeChat
Zhipu AI's desktop ADE puts its open-weight GLM-5.2 model behind a full agent loop with 20+ integrated tools and lets developers steer long-horizon tasks from WeChat, Feishu, or Telegram. Free download; GLM Coding Plan starts at $16.20/month.
Kimi K2.7 Code Is the First Open-Weight Model in GitHub Copilot's Model Picker
GitHub has added Kimi K2.7 Code as a selectable option in Copilot, making it the first open-weight model in the standard model roster. Available now for Pro, Pro+, and Max plans on VS Code; Business and Enterprise require admin opt-in. Copilot CLI also gains automatic task-based model routing on the same day.
OpenAI GeneBench-Pro: Sol Pro Hits 31.5% on Biology Research Tasks — and Accidentally Reveals Three Unannounced Pro Models
OpenAI's new computational biology benchmark puts GPT-5.6 Sol Pro at 31.5% on 129 research-grade genomics tasks, nearly double Anthropic's Opus 4.8 at 16.0%. The same paper quietly names Luna Pro, Terra Pro, and Sol Pro — the first time ChatGPT Pro has split into multiple tiers.
UBTech's UWORLD U1 Ships China's First Mass-Produced Consumer Humanoid in September — 13,361 Pre-Orders at $17,650+
Shenzhen-based UBTech launched three U1 humanoid variants on June 30 priced from 119,800 yuan ($17,650) to 990,000 yuan ($145,000). With 13,361 pre-orders secured and deliveries beginning September 16, it is the first full-size humanoid designed for mass production and home companionship.
Weave Robotics Isaac 1 Ships Fall 2026 at $449/Month — YC Startup Bets Service Model Is Home Robotics' Missing Link
San Francisco-based Weave Robotics opens Isaac 1 pre-orders at a $250 deposit, offering a mobile two-arm home robot at $449/month or $7,999 upfront. California deliveries begin fall 2026, US-wide from 2027.
Anthropic Launches Claude Science: 60-Database Research Workbench Cuts Germline Analysis Time by 90%
Claude Science brings a coordinating AI agent, 60 scientific databases, HPC job submission, and code-traced artifacts into a single session. Early pharma users report germline analyses in one-tenth the time and 10-page review drafts done in bulk.
Anthropic Leads US Enterprise AI Spend for First Time: 41% to OpenAI's 39.5% on Ramp June Index
Ramp's June 2026 AI Index shows Anthropic at 41% of tracked US businesses with paid AI subscriptions, up 2.5 points from May. OpenAI fell 0.1 points to 39.5%. It is the first month Anthropic has led the index.
Researchers Allege Claude Code Embeds Hidden Prompt Markers to Tag China-Linked Proxy Routes
Security researchers claim Claude Code detects non-default ANTHROPIC_BASE_URL routes tied to Chinese operators and silently injects invisible punctuation and date formatting into prompts. Anthropic has not confirmed or denied the claims.
Ramp Tracks 21,559 Firms: Heavy AI Adopters Grew Headcount 10%, Entry-Level Roles Up 12%
A Ramp Economics Lab study linking AI spend data to Revelio Labs workforce records finds that high-intensity AI adopters grew total employment 10.2% and entry-level headcount 12% in the two years after adoption. Low-intensity adopters saw no statistically significant change.