🤖 OpenAI launches GPT-6 Astra with record benchmark scores

🌦️ Google just gave weather forecasts a serious upgrade

💸 Meta cuts AI prices 95% for users who share data

🏭 Volkswagen is cutting 50,000 jobs and half its car models

🚕 Tesla Cybercab is under investigation by the Feds

+ 📊 Daily poll and results

+ 📈 Trending tools and resources

OpenAI introduced GPT-6 Astra, its new flagship model, claiming major gains across science, math, coding, computer use, and cybersecurity, including 99.9% on ARC-AGI-3 versus GPT-5.6 Sol’s 7.8%.

Astra costs $10 per million input tokens and $50 per million output tokens, about 2.5 times GPT-5.6 Sol, while OpenAI says improved efficiency should reduce token use per task.

Despite those benchmark wins, Astra ranks fifth on Artificial Analysis’s Intelligence Index behind four rivals, which makes its launch less of a coronation and more of a very expensive showdown.

Google DeepMind introduced WeatherNext 3, a weather model that refreshes forecasts hourly using live satellite imagery and delivers temperature detail down to five kilometers, outperforming rival AI systems and major weather agencies.

The model combines live satellite feeds and ground-station data instead of relying on physics simulations that arrive about six hours late, cutting rain errors by up to 60% against satellite observations.

Google plans to bring WeatherNext 3 into Search, Maps, Gemini, and Earth, meaning one of AI’s most practical upgrades may quietly show up the next time you check for rain.

Meta is offering a contributor tier for its new Muse Spark agent model that drops input tokens from $1.25 to 10 cents per million, and output tokens from $4.25 to 20 cents.

The discount buys prompts and model outputs for training future models. Meta needs that data badly after pausing an internal employee mouse-tracking program in June following wide criticism inside the company.

Princeton's Arvind Narayanan notes big companies already pay 10x to 20x more for enterprise plans specifically to avoid training. Meta just put a number on what that privacy is worth.

Volkswagen approved its Future Plan 2030, targeting a 9% operating margin by 2030 and cutting 50,000 jobs, including management roles, as it launches what it calls its deepest transformation yet.

The automaker blames tougher global competition, weaker demand, technological shifts, U.S. tariffs, and lost sales in China, while its European factories currently have capacity for 500,000 more cars than demand requires.

Volkswagen also plans to halve its model lineup and may repurpose or close European plants, a sign that survival now means making fewer cars, in fewer variants, with fewer people.

NHTSA opened an investigation hours after Tesla launched Cybercabs on Austin’s public roads without steering wheels or pedals, questioning how the company certified vehicles that appear to conflict with current federal safety rules.

Tesla told NHTSA it self-certified the Cybercab under existing vehicle standards, while regulators now want the technical basis for that decision, including which manual-control requirements Tesla considers inapplicable.

Tesla is not the first to test this regulatory shortcut: Zoox tried similar self-certification in 2022, then spent years securing exemptions before launching paid robotaxi rides.

A) Yes - 26%

B) No - 74% 🏆