RuntimeWire

Real-time startup intelligence for the AI economy. Funding, model launches, infra shifts, and founder moves — the signals that matter.

Latest stories — page 4

• Anthropic weighs shareholder sales in IPO and a lockup longer than 180 days

  The AI developer plans to publish its prospectus after Labor Day, ahead of a possible late September or early October listing.

• FDA authorizes Vitestro's Aletta robot for autonomous blood draws

  Aletta handles the needle stick, tube changes and bandaging, while one trained phlebotomist can supervise three machines.

• DeepSQL's benchmark finds BRIN's 5% update cliff in Postgres

  Venkat Sakamuri's 10M-row test found a 48 kB BRIN index touching 28 times more heap pages as cumulative updates rose from 1% to 5%.

• Predoc launches Curated Data to turn faxed charts into AI-ready records

  The New York startup is extending its retrieval service into a normalized patient-data layer delivered through an app or API.

• Forecast Bio discloses $2.4M SAFE financing for its astrocyte AI platform

  Forecast Bio's filing covers nine investors, with the first sale dating to November 2025 and $750,000 of the offering still open.

• Metriport raises $26M to turn health-data access into clinical answers

  Matrix's TJ Parker led the round as the founders expand an open-source system that retrieves, cleans and interprets records inside clinical workflows.

• Quaise Energy closes $180M Series B to commercialize superhot geothermal

  Quaise Energy will use the round to move millimeter-wave drilling toward commercial revenue and a 50 MW first phase at Project Obsidian by 2030.

• Colin Percival launches Route 53 Files, an unofficial AWS DNS filesystem

  The Tarsnap founder built a free service that lets UNIX commands edit Route 53 records, then documented the sharp edges in AWS's own voice.

• Socure raises $156M, buys Fravity to automate fraud investigations

  Johnny Ayers' identity platform is valued at $5.2B as it folds two PayPal fraud veterans' investigation software into RiskOS.

• Google DeepMind pilots double-blind AI evaluations without sharing prompts or model weights

  Google DeepMind says its pilot keeps external evaluation prompts and model weights private, though its first-of-its-kind claim remains independently unverified.

• HeyGen details TAVR, a video-reference system for steadier AI avatars

  First disclosed on March 3rd, the production research system uses up to 48 reference frames to preserve a person's identity across new scenes.

• OpenAI hid Asteroids, Snake and Brick Breaker inside the Codex Micro

  OpenAI's $230 controller turns recent assistant messages into the asteroids, bricks and food inside three undocumented games.

• Mara raises $7M to make counter-drone defense as expendable as the threat

  Daniel Kofman and Sriram Raghu built Spike around tiny kinetic interceptors, with company-reported tests in Army exercises and Ukraine.

• Merck co-leads Faro AI's $37.3M bet on structured clinical-trial agents

  Scott Chetham built Faro after seeing static protocols slow trials; the new round funds agents that can carry study intent into downstream workflows.

• Z.ai will release GLM-5.3 weights on August 28th after safety delay

  The Tsinghua-born AI lab held back the coding model's checkpoints for two weeks after its cyber capabilities grew faster than expected.

• OpenAI built ChatGPT interfaces that keep running after answers appear

  Codex client code reveals a server-directed GenUI lane that can refresh specific widgets after delivery, separate from OpenAI's public Apps SDK for third-party MCP apps.

• xAI puts Grok 4.6 on Microsoft Foundry after AWS and Google Cloud

  The public preview gives Azure customers access to Grok 4.6 with a 500,000-token context window, extending xAI's multi-cloud distribution push.

• Nvidia agrees to buy Hugging Face for $12.9B, report says

  The deal gives Nvidia control of a central open-model distribution layer at roughly 86 times Hugging Face's annualized revenue.

• Claude Opus 5 (Fast) tops Newsroom Reliability v0.2 benchmark

  In a 50-task run of Newsroom Reliability v0.2, Claude Opus 5 (Fast) led the field with a 0.73 score at $0.0460 per task. GLM 5.3 Flash, Gemini 3.7 Flash, and Qwen3.8 Flash followed at 0.69, each at substantially lower cost.

• SpaceXAI tests voice calls for Grok Bot's persistent AI workers

  A test interface shows users speaking directly with agents that already browse websites, operate software and run jobs in the cloud.

• Z.ai GLM 5.3 Flash tops Editorial Craft benchmark at 0.94

  In a 12-task Editorial Craft evaluation, Z.ai’s GLM 5.3 Flash ranked first with a 0.94 score and an estimated cost of $0.0002 per task. Step-3.7-flash followed at 0.92, while Qwen3.8 Flash, Claude Opus 5 (Fast), Gemini 3.7 Flash, and DeepSeek-V4-Flash-0731 rounded out the leaderboard.

• Sword Health plans a $300M Headspace buyout after its $3B pandemic mark

  Debt and roughly $397M in disclosed equity funding leave common shareholders at the back of a crowded payout line.

• Imageat serves unidentified uncensored Qwen derivative, leaving customers to test it

  The third-party provider offers metered access without identifying the checkpoint, revision, modification method or endpoint evaluation behind its service.

• Army selects Antares to build and operate Fort Bragg microreactors in $2.2B Janus program

  Jordan Bramble's defense-first nuclear startup will own and operate the reactors, with payments tied to technical milestones and operation targeted for 2028.