- 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 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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%.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.