Morning Edition · Wednesday, September 23, 2026
Tech
OpenAI Halves API Prices With GPT-6 Sol and Luna as Anthropic Cuts Opus Running Costs 40%
Both releases came within hours of each other on September 22. OpenAI priced its cheaper model, Luna, at 10 cents per million input tokens, and Anthropic says Opus 5.5 matches Fable 5.1-level work at a lower cost.
Tech
Epoch AI Puts the Fall in the Cost of a Benchmark Answer at 725-Fold in Under 18 Months
OpenAI's o3 needed about 30 cents per question to reach 75 percent on the graduate-level science benchmark GPQA Diamond. GPT-5.6 Luna matched that score for four hundredths of a cent.
Geopolitics
China's Internet Regulator Opens Investigation Into DeepSeek and Moonshot Over Data Routed to Claude
Anthropic alleges that Moonshot sent more than 23 million exchanges to Claude between May and July, and that DeepSeek sent more than 12.1 million in a 14-day window. The allegations raise the question of whether Chinese user data reached United States (US) servers.
Geopolitics
Trump Tells the United Nations the US Will Call Artificial Intelligence "Super Intelligence"
The president paired the renaming of the technology in federal documents with a rejection of any global mechanism to control it. The White House has not said how the change will be implemented.
Tech
OpenAI Forms a Mathematics Advisory Group at the Institute for Advanced Study After Claiming 100 Solved Open Problems
Nine mathematicians, including Timothy Gowers and Edward Witten, will assess the significance of results, but OpenAI says the group will have no say over the pace of its internal research.
Tech
A StarCraft Benchmark Finds Agents Can Win Matches Without Learning Strategy
Across 171 round-robin games, a Codex harness running GPT-6 Astra went 18-0 at the highest reasoning setting, yet the benchmark's author says no tested system beat a basic human player.
Tech
NVIDIA Ships Isaac ROS 5.0 With Agent Workflows for the Robot Operating System Ecosystem
The release, announced at the ROSCon conference in Toronto, adds agentic development tooling and claims object tracking up to 5.5 times faster, targeting roughly 1.3 million users of the open framework.
Tech
OpenAI Publishes Rules for Outside Safety Assessors as Protesters Camp Outside Its Offices
The company names four areas it wants third parties to test and seven principles for how, while a daily protest that began in San Francisco in July has moved to its New York office.
Tech
Penn Researchers Combine Agentic Reasoning With Geometry Models to Build 3D Scenes From a Single Image
The HARMONY framework uses a hierarchical chain of thought to decompose a photograph before handing geometry to a visual foundation model, an approach that treats scene reconstruction as a planning problem.
Tech
The Choice of 4-Bit Quantizer Decides Whether a Fine-Tuned Model Leaks Private Data
Researchers found that calibration-free GGUF Q4_K_M reproduced 5.3 percent of planted private records while calibration-based AWQ and GPTQ reproduced none, at the same bit width.
Macro
Anthropic's Economic Model Puts 2030 US Output Between $34.1 Trillion and $44.4 Trillion Depending on AI Adoption
In the most aggressive of three scenarios, annual growth reaches 15 percent while unemployment among knowledge workers climbs to roughly 18 percent against under 4 percent for everyone else.
Tech
NVIDIA and OpenAI Push Southeast Asian Deployment With Singapore Event and a 30,000-Partner Training Program
NVIDIA held a two-day developer event in Singapore while OpenAI and the ride-hailing company Grab launched a regional skills program, both aimed at the deployment layer rather than model capability.