RuntimeWire

Featured

• OpenAI quietly ships "Watch and fix" for Codex pull requests

  The production desktop app can assign Codex to keep checking a PR every 10 minutes, repair CI failures and merge conflicts, push fixes, and optionally continue until the pull request merges.

• OpenAI prices GPT-6 Astra at Claude Fable 5.1's $10/$50 rate

  The new model costs 2.5x more per token than GPT-5.6 Sol's current promotional rate, putting the premium on autonomous work.

• Hidden ChatGPT modal previews new ImageGen 2.5 model

  The unreleased interface promises higher-quality results and smarter tools, while OpenAI's public documentation still lists Images 2.0.

• OpenAI releases GPT-6 Astra as Brockman declares the 'AGI era' has begun

  The flagship costs 2.5 times more than GPT-5.6 Sol, reaches cyber customers first and produces reasoning OpenAI says is harder to monitor.

• Codex leak labels OpenAI's Astra as GPT-6 ahead of release

  Two model IDs appeared in a Codex Desktop feature flag two days after OpenAI said Astra would be available soon.

• Lyte raises $165M to build the sensor stack robots keep rebuilding

  Maverick Silicon led the round at a $1.6B valuation, backing Apple and PrimeSense veterans Alexander Shpunt, Arman Hajati and Yuval Gerson.

• MiniMax H3 generates a 10.125-second audiovisual file in under nine seconds

  MiniMax's open-weight H3 now runs through vLLM-Omni, while its community license excludes the US, EU, UK and Republic of Korea.

• OpenAI is turning Flipper’s BUSY Bar into a physical ChatGPT status light

  RuntimeWire found a hidden, disabled-by-default BUSY Bar control in the latest desktop build, plus a dedicated animation package for task, ready, pending, error and disconnected states.

Investigations

• OpenAI quietly ships "Watch and fix" for Codex pull requests

  The production desktop app can assign Codex to keep checking a PR every 10 minutes, repair CI failures and merge conflicts, push fixes, and optionally continue until the pull request merges.

• OpenAI is turning Flipper’s BUSY Bar into a physical ChatGPT status light

  RuntimeWire found a hidden, disabled-by-default BUSY Bar control in the latest desktop build, plus a dedicated animation package for task, ready, pending, error and disconnected states.

• Kimi Work 3.2.4's hidden system prompt override triggered local file reads without approval

  A controlled Windows test reproduced the read in Manual approval mode; a restored-default control stayed clean, and no network transmission occurred.

• Anthropic asks court to restrict Reddit’s AI team from Claude’s trade secrets

  Anthropic says Reddit's internal AI work makes some in-house counsel a competitive risk during discovery.

• OpenAI built a hidden task system for ChatGPT inside Excel and PowerPoint files

  The unreleased desktop feature turns comments on cells, charts, slides and shapes into threaded assignments that ChatGPT can complete in batches.

• Kimi is building a PopClip-style AI toolbar for selected desktop text

  Code in Kimi Work 3.2.3 detects highlighted text systemwide, but its search, translation and summary buttons are still unwired.

View all investigations

Stories

• Atira raises $17.5M after landing industrial customers without a sales team

  Accel led the $15M seed as Palantir and Google alumni Florian Diegruber and August DuMont Schuette prepare Atira's first commercial hires.

  0 upvotes

• Figure commits $3.5B in compute to Nscale for up to 100,000 Rubin GPUs

  Figure targets initial deployment in Barstow, Texas, in the second half of 2027; Nscale will take an undisclosed stake and serve as Figure's preferred compute provider.

  0 upvotes

• OpenAI releases GPT-6 Astra as Brockman declares the 'AGI era' has begun

  The flagship costs 2.5 times more than GPT-5.6 Sol, reaches cyber customers first and produces reasoning OpenAI says is harder to monitor.

  3 upvotes

• Fal launches H3 Max Director for live, steerable AI video

  Fal's H3 Max Director turns short generated clips into a WebRTC session that developers can steer with live prompts. The API supports 24 fps output and remembers up to 50 earlier prompt segments. Fal launched it with a [75% generation discount](https://x.com/fal/status/2095599871449342288) for the first two weeks.

  0 upvotes

• Codex leak labels OpenAI's Astra as GPT-6 ahead of release

  Two model IDs appeared in a Codex Desktop feature flag two days after OpenAI said Astra would be available soon.

  3 upvotes

• Resect AI raises $25M to inspect and modify enterprise LLM behavior

  Kevin Owens and three co-founders are betting enterprise buyers will pay for model-level intervention beyond dashboards that score bad answers.

  0 upvotes

• Kacey Musgraves puts Meta Quest viewers inside a Texas honky-tonk

  The VR180 production moves viewers from stage to guitar circle to rodeo, making the headset part of the concert rather than another screen.

  0 upvotes

• Osaurus ships a 6.3 GB agent model for 8GB and 16GB Macs

  Raptor 0.5 activates about 1B of 7.9B parameters and targets file, spreadsheet and email tasks on Apple Silicon.

  2 upvotes

• SerenX discloses $2.3M offering a year after first sale

  Scott Hasbrouck's Florida defense startup sold the offering to six accredited investors and says its 2.8-kilogram interceptor reached Technology Readiness Level 6.

  2 upvotes

• Cloudflare launches OpenAI-powered vulnerability defense tied to live traffic

  Matthew Prince's latest security bet uses GPT-5.6 Cyber to rank findings, propose patches and prepare WAF rules, with humans approving every change.

  0 upvotes

• OpenAI prices GPT-6 Astra at Claude Fable 5.1's $10/$50 rate

  The new model costs 2.5x more per token than GPT-5.6 Sol's current promotional rate, putting the premium on autonomous work.

  1 upvote

• Crusoe signs reported $13B Jane Street cloud deal amid $3B fundraising talks

  The five-year GPU contract would turn a former quant trader's energy startup into Jane Street's second multibillion-dollar AI cloud supplier.

  0 upvotes

• NVIDIA PAIR turns spare PCs into a local AI inference cluster

  The open-source beta routes parallel agent requests across Windows, Linux and macOS devices without pooling their GPU memory.

  0 upvotes

• TrndX raises $3.175M to make an AI cloud fit in a rack

  The fully sold equity offering gives optical-switching veteran Jitender Miglani fresh capital for TrndX's UniFlex infrastructure platform.

  0 upvotes

• Browser Use gives web agents single-use cards for checkout

  The Link integration lets users approve purchases without exposing their underlying card details to the browser agent.

  1 upvote

• Google DeepMind plans to bring WeatherNext 3 forecasts to Search, Maps and Gemini

  The joint Google DeepMind and Google Research project is moving into consumer products, developer APIs and cloud datasets, where company-reported benchmark gains will meet real weather.

  1 upvote

• Google rolls Gemini voice controls into Gmail, Docs and Keep

  Gmail Live and Keep Live reach Plus, Pro and Ultra plans; Docs Live requires Pro or Ultra, with business access promised later.

  1 upvote

• IFM releases six K2 Horizon models with training records for outsiders to inspect

  Eric Xing's MBZUAI institute published models from 0.9B to 375B parameters, along with code, data recipes, checkpoints and benchmark caveats.

  1 upvote

• Ansh Chokshi and Shashwat Kapoor build Mireye's location layer for AI agents

  Mireye's API and MCP server return cited parcel, hazard, utility and site-selection data across the US.

  1 upvote

• MiniMax M3 powers HUMAIN's 428B-parameter Arabic model

  HUMAIN further trained MiniMax M3 on more than 1 trillion Arabic tokens and put the 428-billion-parameter result into a limited API preview.

  2 upvotes

• Context7 launches Docs7, a Mintlify-compatible documentation host built for agents

  Upstash's new product serves Markdown, WebMCP search and agent analytics, with a free tier and a $200 monthly Pro plan.

  0 upvotes

• Botika runs a 100-terabyte fashion AI stack on Modal, minus Kubernetes

  Eran Dagan says moving data, training and 15 production models onto Modal let Botika scale research without doubling its infrastructure staff.

  2 upvotes

• Thinking Machines seeks $1B-plus at a $40B valuation after last fall's $50B ask

  The lower target still values Mira Murati's AI lab at more than three times its July 2025 post-money price.

  0 upvotes

• Hidden ChatGPT modal previews new ImageGen 2.5 model

  The unreleased interface promises higher-quality results and smarter tools, while OpenAI's public documentation still lists Images 2.0.

  0 upvotes

• Zeit AI raises €5M to bring an autonomous data engineer to Europe's mid-market

  The Palantir alumni behind ZeitMind say they reached €1M ARR in 12 months; YC, Oxford's Seed Fund and Sequoia scouts joined the round.

  2 upvotes

• Tesla opens Cybercab fleet interest list, leaves the economics for later

  The form asks about fleet purchases and deployment regions but gives no price, delivery schedule or Robotaxi revenue-sharing terms.

  0 upvotes

• Polars 2.0 makes streaming automatic and implicit row order explicit

  Ritchie Vink's release candidate lowers memory demands while tightening the rules around ordering, type coercion and malformed data.

  2 upvotes

• Souk raises $1.6M to turn dormant B2B partnerships into sales

  Souk's Leo Crowe, Sofia Hamilton Sambrano and Ayooluwa Alfonso are building Coco, an AI agent designed to reactivate dormant business partnerships and connect them to sales. Sure Valley Ventures led the pre-seed round, with Antler and Fuel Ventures participating.

  2 upvotes

• Guardio raises $40M at $1.1B, keeping dilution on a short leash

  Amos Peled says Guardio has passed $150M in ARR and 1M paying customers after bringing Wiz co-founder Assaf Rappaport into the round.

  1 upvote

• Flock Safety's AI watchlists let police search for people by description

  A reconstructed interface shows map-based person searches, image-ranking feedback, and moderation controls that can block or warn on sensitive queries.

  1 upvote

• OpenAI teases Astra with MIT's 1979 voice-and-gesture interface

  The clip points toward a multimodal launch two days after OpenAI said its delayed Astra model would arrive soon.

  0 upvotes

• Anthropic cuts Fable 5.1 cache-read prices 75% for long-running agents

  Input remains $10 and output $50 per million tokens, while Anthropic estimates the cache discount can lower highly agentic workload costs by roughly 45%.

  0 upvotes

• OpenAI investigates ChatGPT and Codex outage across nearly every major feature

  Errors hit conversations, login, Search, uploads, Voice, Agent and Codex; OpenAI had not published a cause.

  0 upvotes

• NTU researchers release Puffin-World with camera-grounded 3D states

  According to the [Hugging Face model card](https://huggingface.co/ACERobotics/Puffin-World?ref=runtimewire), the NTU-led model generates camera-controlled RGB and depth views that can be consolidated into 3D point clouds.

  1 upvote

• Qdrant releases a 10-billion-vector benchmark and an engine to rerun it

  FineWeb-10B contains about 10.07B vectors; Supernova turns Qdrant's giant test into an open workflow for evaluating rival databases.

  1 upvote

• Modal opens London office to sell its AI cloud across Europe

  Erik Bernhardsson is pairing a new commercial base with the Stockholm engineers who built core parts of Modal's infrastructure.

  0 upvotes

View all Stories