RuntimeWire

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

Latest stories — page 13

• DeepSeek's experimental vision model spans three formats, caps images at 384 tokens

  Liang Wenfeng's separately named `deepseek-v4-flash-vision-exp` supports Chat Completions, Messages and Responses requests, with each image billed at no more than 384 V4-Flash tokens.

• KidsHustle's 14-year-old founder seeks £500,000 for teen jobs marketplace

  Sasha Bagrov is raising pre-seed capital to expand an iOS marketplace built around parent approvals, verified job owners, escrow payments and live location sharing.

• SGLang publishes one-GPU Qwen3.8-27B recipes, claims 206.1 tokens per second

  Ying Sheng and Banghua Zhu's inference project added NVFP4 and DFlash2 recipes, with project-reported throughput of 206.1 tokens per second on one RTX 5090.

• LibAI Lab launches NewFace to turn product photos into ad campaigns

  NewFace combines a node-based canvas, reusable ad formats and third-party video models for sellers and agencies.

• Genki puts a screen on its $279 Manta controller so the app can stay home

  Eddie Tsai's customizable gamepad starts at $189 on Kickstarter, with TMR sticks, analog buttons and profiles managed on-device.

• Twitch opens sponsorships to Affiliates, with Wehype supplying campaigns

  EA's Battlefield is the first named Wehype offer, while certification and brand criteria keep paid work from being automatic.

• RoboStore ends Unitree distribution and bets on US robot assembly

  Teddy Haggerty's Robo Inc. plans a 66,000-square-foot Long Island plant after FCC restrictions blocked new foreign-produced robot models from receiving the US equipment authorization required for import or sale.

• Anthropic's $1.5B Ode builds an AI consultancy around private equity portfolios

  Chris Taylor's 100-engineer operation gives Claude a deployment arm and its financial backers a ready-made customer pipeline.

• Pew finds AI fingerprints on 35% of newer webpages it could date

  The 490,000-page study found a 10% rate overall, with commercial sites far ahead of government and education domains.

• OpenCode says its $10 Go plan passed 301,000 active subscribers

  A dashboard shared by founder Jay V showed 301,018 active accounts and 14.6% growth, giving OpenCode a sizable paid base beyond its free agent.

• AliExpress anti-bot scripts blocked one user's Bluetooth handoff by playing nothing

  Two obfuscated anti-bot scripts opened live audio graphs, exposing a gap in browser mute controls and producing an unusual hardware side effect.

• Apple Music will label AI-made tracks, artwork and videos later this year

  Providers must flag material AI use across recordings, compositions, artwork and videos, while Apple's public delivery spec still calls the metadata optional.

• ChatGPT can now search and send Apple Messages from your Mac

  The plugin is available in ChatGPT Work and Codex on desktop, with message access and sending governed by app permissions.

• Locus launches one balance for 600 APIs, because agents collect subscriptions too

  Cole Dermott's YC-backed startup is selling platforms a metering layer spanning models, search, scraping and data providers.

• Agility Robotics targets fall SPAC close after Unitree's 460% debut

  Peggy Johnson is pitching industrial deployments and $300M in potential multi-year contract value as Agility seeks a $2.5B pre-money valuation.

• Liquid AI trains 4-bit LFM2.5 checkpoints to retain roughly 97% of BF16 performance

  The MIT spinout used quantization-aware distillation across four LFM2.5 models, then tested them on a phone, laptop, mini PC and Raspberry Pi.

• GPT-5.6 Sol Pro tops Editorial Craft benchmark with 0.97 score

  In a 12-task Editorial Craft evaluation of 12 AI models, OpenAI’s GPT-5.6 Sol Pro ranked first with a 0.97 score at $0.0091 per task. OpenAI claimed four of the top six spots, while Grok 4.6 and Claude Opus 4.8 tied at 0.93 among the leading non-OpenAI models.

• GitHub blames capacity failure and retry storm for nearly eight-hour outage

  A Central US autoscaling failure hit authentication, Actions and Copilot as GitHub races to move production traffic onto Azure.

• SpaceXAI opens Grok Build to every plan and gives its apps an X feed

  SpaceXAI's app builder now spans web and mobile, publishes to grok.me and lets projects call Grok models without separate API keys.

• Micro1's Ali Ansari offers $12.5M for Spirit's data after Google's auction win

  The late bid is 25% above Google's, but Spirit's auction record shows privacy terms may matter as much as the extra cash.

• Daniel Vaughn publishes Huzzah, an AI editor built around persistent pseudocode

  The experimental editor treats persistent `.hz` pseudocode as a human-authored specification and regenerates source code from changes to it.

• YipitData explores a sale above $2.5B after its Groupon-era pivot

  Goldman Sachs is testing buyer interest as the data provider targets about $280M in 2026 recurring revenue, Reuters reports.

• Anthropic launches Claude Academy, turning employee training into product onboarding

  The free catalog mixes general AI fluency courses with tutorials for Claude.ai, the Cowork agent, Claude Code, Claude Tag and the API.

• Attackers poisoned three Rust crates to steal developer credentials during builds

  The malicious releases remained online for 86 to 107 minutes across packages with roughly 264M lifetime downloads.