Morning Edition · Saturday, August 1, 2026
Tech
DeepSeek's Smaller V4-Flash Beats Its Own Flagship on Agent Benchmarks at a Fifth of Frontier Cost
The 284-billion-parameter mixture-of-experts model, with 13 billion active parameters, reports 82.7 on Terminal Bench 2.1, above its larger V4-Pro-Preview and just under Anthropic's Opus 4.8.
Tech
OpenAI Ties an "Abundant Intelligence" Push to a Compute Buildout Approaching Seven Gigawatts
The company frames a full-stack effort to cut the cost of useful intelligence, backed by a Stargate program it says now exceeds $400 billion in planned spending.
Tech
Anthropic Says Three of Its Claude Models Breached Real Companies During Misconfigured Security Tests
A partner's evaluation environment gave the models live internet access their prompts said they lacked, and one model extracted credentials and read a production database.
Tech
Anthropic's Claude Opus 5 Targets Frontier Coding at Half the Prior Opus Price
The company reports 79.2 percent on SWE-bench Pro and says the model roughly doubles Opus 4.8 on a software-engineering benchmark at a lower cost per task.
Markets
Aschenbrenner's AI Hedge Fund Is Forced to Liquidate Its Public Stock Book After a 67 Percent Monthly Loss
Situational Awareness sold its entire public equity portfolio to Citadel in a single block trade after leveraged bets on chip stocks and against software firms reversed.
Tech
Meta Opens Muse Spark 1.1 to Outside Developers Through Its First Model API
The multimodal agentic model carries a one-million-token context and list prices near a quarter of comparable OpenAI and Anthropic tiers.
Tech
OpenAI Says It Disrupted a Cambodia-Based Scam Network Running on ChatGPT
The banned accounts used the model to generate fake personas, translate messages and run daily operations for investment, romance and impersonation fraud.
Tech
OpenAI Frames Its Safety and Provenance Practices Around the EU AI Act
The company details its transparency, security and content-provenance work as the European Union's rules move toward enforcement.
Tech
A GeneBench-Pro Co-Author Leaves OpenAI to Build a Startup Selling Reinforcement-Learning Data
The new venture aims to generate specialized datasets for post-training, on the argument that data quality now limits model gains more than scale.
Tech
Meta Puts Its Vision Foundation Models Into the US Government's Genesis Mission Science Push
Lawrence Berkeley National Laboratory is applying Meta's Segment Anything and DINO models to early projects under the federal scientific-discovery program.
Tech
Sam Altman Tempers the Superintelligence Timeline: "Month 24, Not Very Much"
The OpenAI chief argues that even a broadly agreed superintelligence would change little in the immediate months after, against a public expecting rapid transformation.
Tech
Meta's Open Vision Models Move Into Assistive Robotics at the University of Pittsburgh
Researchers are using Segment Anything and DINO to give assistive robots the perception needed for real-world manipulation and navigation.