RuntimeWire

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

Latest stories — page 12

• Munder Difflin fixes agent cost, memory and messaging failures in v0.4.5

  Chaitanya Giri's local-first harness wraps 12 agent CLIs. Its v0.4.5 release notes document repairs to cost reporting, Apple Silicon semantic memory and communication between workers.

• Dan Luu reports a 7% ripgrep speedup after minutes of prompting

  Dan Luu argues coding agents have made once-specialized optimization cheap enough to customize software for each workload.

• Outer Biosciences has raised about $23M to build a skin-assay licensing business

  Outer Biosciences, Michael Polansky's 19-person company, uses machine learning and human skin tissue that remains viable for four weeks to find compounds for beauty and pharmaceutical companies.

• Darius Monsef's OzBrain gives AI agents one shared memory

  The 3x YC founder built a staged, Markdown-based knowledge layer for Claude, ChatGPT and Cursor, priced from free to $99 per month.

• Cloudflare syncs AI bot rules with robots.txt, closing its own policy gap

  Product manager Jin-Hee Lee's feature is headed for an upcoming rollout, publishing Search, Agent and Training choices while preserving existing Disallow rules.

• Base Compute gives AI agents the kernel-tuning job for open models

  Its B:OS demo took Nemotron 3 Nano from unsupported to 1.43x faster decode than MLX on Apple silicon, by Base Compute's tests.

• Tumble Forth rebuilds the programming stack from bare metal to C

  Virgil Dupras's 11-chapter tutorial turns his collapse-minded OS work into a path beneath the modern software toolchain.

• Rillet raises $100M at a $1B valuation to replace accounting software

  ICONIQ led the Series C after Rillet reported that its annualized revenue rate had doubled in one quarter, though the company did not disclose its underlying revenue.

• SpaceXAI puts Grok 4.6 in Google Cloud's Vertex AI Model Garden

  Google's August 21st release notes list the model as a Preview offering, while SpaceXAI quotes $0.30 cached input and a 500,000-token context window.

• MiniMax hints at M3 on SambaNova, provides no benchmark result

  An undated MiniMax post mentions M3 and SambaNova hardware, while the companies' disclosed demonstration identifies M2.7 and provides no M3 setup or result.

• OpenAI cuts GPT-5.6 Sol output pricing by one-third for three months

  The flagship model now costs $4 per million input tokens and $20 per million output tokens, while paid-plan usage limits stay put.

• Grok declared a Chinese influence campaign real. Ten seconds after a ChatGPT audit, it backed down.

  After showing a 120-source counter, Grok treated a disputed $23.6B data-center tally as evidence of a foreign-directed US campaign. ChatGPT found that the underlying sources did not establish the money, instructions or control required for that conclusion.

• Khosla backs Jeff Dean's plan to automate science, starting with AI research

  Discovery Loop has announced an undisclosed initial round co-led by Khosla Ventures and Radical Ventures as Jeff Dean and three Google veterans build a machine-learning research system first.

• Shanghai AI Lab published an August 13th paper on its 397B scientific-agent model

  Intern-S2-Preview combines visual pretraining, tool use and long-horizon reinforcement learning in an open 397B-parameter scientific model.

• Grok Bot opens to $60 Cursor Pro+ and a limited free trial

  Access now includes SuperGrok Plus, Cursor Pro+, and all Cursor Teams plans, while every other user gets a usage-limited trial.

• NVIDIA's AVO scores 100% on ARC-AGI-3's public set

  Powered by Claude Opus 5, the coding-agent system cleared 183 levels in 6,624 actions, though NVIDIA did not test the private sets.

• Anthropic puts Claude Security's GitHub scanner behind an Enterprise login

  Claude Security entered public beta for Claude Enterprise customers in late April 2026; Anthropic keeps the more capable Mythos 5 in a separate trusted-access program.

• DJI veteran Brett Wang launches Morpho UV printer, says pledges hit $7M

  Shenzhen-based Morpho reported nearly 2,000 backers in 16 hours; Morpho Studio is operational, though custom-jig features and a roll-to-roll UV-DTF attachment remain in development.

• Moonshot released Kimi K3's weights on July 27th, extending its open-weight push

  Moonshot launched the 2.8-trillion-parameter model on July 16th and released its weights 11 days later, pairing a one-million-token context window with CEO Yang Zhilin's long-running research thesis.

• Z.ai offers up to 5 trillion GLM-5.3 tokens to recruit ZCode users

  The two-day promotion gives 50,000 new users 100M tokens each, extending a coding-agent push that Z.ai says already reached 1M users.

• Starcloud raises $250M as orbital AI runs short of rockets

  Manhattan West led the extension at a $2.3B valuation, giving Philip Johnston capital for a factory, spacecraft and scarce launch slots.

• MyDataWork puts a $7,500 analytics context plan on AWS Marketplace

  Founder Gib Bassett is betting that easier cloud procurement can move his metadata workspace from analyst trials into shared team deployments.

• Argentic bets AI scrapers will pay 10 sats for a proxy hop

  The pseudonymous builder said the zero-cost prototype had no paying agents after seven weeks, a candid test of whether bots will pay instead of route around.

• VLM Run puts six open OCR models behind one OpenAI-compatible endpoint

  Founder Sudeep Pillai is betting orchestration will matter longer than any single document model.