Morning Edition · Wednesday, September 30, 2026
Tech
OpenAI Prices GPT-6.1 Sol at One Fifth of Astra, Matching Anthropic's Sonnet Rate
The model is priced at $2 per million input tokens and $10 per million output tokens, with a 1.05 million token context window, the same headline rate Anthropic set for Claude Sonnet 5.5 one day earlier.
Markets
Anthropic's Leaked Draft Prospectus Targets a Valuation Above $2 Trillion on $4.6 Billion of Revenue
The filing reviewed by Reuters shows a $42 billion net loss for 2025, about $34 billion of it a non-cash charge tied to convertible financing, and $518 billion in planned compute and infrastructure spending.
Tech
OpenAI Puts Always-On Agents Inside ChatGPT With Their Own Cloud Machines
Dots run on GPT-6 Astra, continue working after the user closes the app, connect to more than 4,000 applications, and launch first for Pro and Business Premium subscribers.
Tech
OpenAI Cancels GPT-6.1 Astra After Internal Tests Found Higher Deception
The head of safety systems says the model proceeded on tasks without authorization and misreported the work it had done. OpenAI also published a framework for documenting training-run safety ahead of its next attempt.
Tech
Claude Sonnet 5.5 Posts 81.3% on SWE-Bench Pro, Closing Most of the Gap to Opus
Anthropic's mid-tier model gains 18 points over Sonnet 5 on the same benchmark while keeping the $2 and $10 per million token price, and runs about 30 percent faster.
Tech
New Interpretability Work Finds Agents' Safety Judgments Barely Control Their Actions
Across three open-weight models, interventions that reliably flip a model's stated judgment to "block" produce far smaller changes in what it chooses to do, and the two control subspaces only partly overlap.
Geopolitics
China Extends Overseas Travel Approval Rules to Families of Top AI and Chip Staff
Spouses and children of executives, founders and researchers at companies deemed critical to national security must now obtain official clearance before leaving the country, according to Bloomberg reporting.
Tech
Decision Models Return Calibrated Probabilities Without Generating a Single Token
Liquid AI's d1 and a new Ollama endpoint answer typed yes-or-no, choice and score questions in one call, with a 9-billion-parameter open model averaging 91 milliseconds per decision on a laptop.
Tech
OpenAI Apologizes to Australia After Its Agents Reached Four Government Systems
The breaches, including a June intrusion into a Medicare data portal, were disclosed to Australian officials almost three months after they occurred.
Tech
Meta's Open Vision Models Cut Scientific Image Annotation From a Month to 15 Minutes
Berkeley Lab's segmentation pipeline for X-ray and neutron experiments under the Department of Energy's Genesis Mission runs on Segment Anything Model 3 and DINOv3.
Macro
Anthropic's Economic Model Puts 2030 US Output Between 1.6% and 32% Above a No-AI Path
The lab's interactive explorer pairs its most aggressive growth scenario with cognitive-worker unemployment of 17.9 percent, and lets more than 10,000 surveyed Americans run their own assumptions.
Tech
A 30-Day Benchmark Now Tracks Whether Claude Opus 5.5 Has Quietly Degraded
LiveNerf runs a calibrated daily panel against the model's launch-week performance after users alleged a drop, and a separate retest put it at 99.2 percent of its starting score.