Weekly Intel - 2026-09-27
OpenAI and Anthropic both shipped models that bring frontier-level performance down in price. At the same time, courts and juries spent the week ruling on whether the companies behind these systems can be trusted, from the Pentagon’s Anthropic ban to Facebook’s Cambridge Analytica verdict to unsealed filings in the Authors Guild case against OpenAI.
AI & Software
GPT-6 Sol and Luna OpenAI added two models to its GPT-6 lineup, Sol and Luna, following GPT-6 Astra, its most capable model. Sol and Luna were trained with methods similar to Astra’s and bring its gains in professional work, factuality, coding, computer use and alignment at higher speed and lower cost. OpenAI says caching and inference improvements cut its serving costs and it is passing the savings to customers, with Sol priced at $2 per million input tokens and $10 per million output tokens.
Claude Opus 5.5 Anthropic released Claude Opus 5.5, the first model in its Claude 5.5 family. It performs at the level of Claude Fable 5.1 on most work, and Anthropic says it costs 40% less than Opus 5 on typical workloads at default settings. External evaluators including Frontier Design and METR tested it before release, and it scored highest of any model on Anthropic’s automated behavioral audit. Artificial Analysis scores it 58 on its Intelligence Index at maximum effort, against a median of 26, with pricing of $4 per million input tokens and $20 per million output tokens.
Tokens too cheap to meter An essay collecting cost data across proprietary, open weight, hosted and local models estimates that the cost of running LLMs fell by about 2.5 orders of magnitude over the past year, mostly from models getting more efficient per task. The author projects LLMs becoming embedded infrastructure within a year or two, and frontier-quality models running locally on commodity hardware within three to six years.
Science
Claude discovers a novel enzyme system with CRISPR-like repeats Anthropic launched a life sciences research group and laboratory that uses Claude for fundamental biology, including searching DNA datasets for uncharacterized protein families and testing the resulting hypotheses in the lab. In an early result, Claude found a previously unknown enzyme system with properties similar to CRISPR, working from high-level direction from Anthropic’s scientists.
Legal & Regulation
U.S. appeals court upholds designation of Anthropic as supply chain risk A federal appeals court panel in Washington, D.C. voted 2-1 on Friday to uphold the Pentagon’s blacklisting of Anthropic, rejecting the company’s argument that the designation was arbitrary, unauthorized and unconstitutional. Judge Gregory Katsas wrote for the majority that the Defense Department had ample support for concluding that continued use of Claude in its information systems posed a national-security risk, with Judge Karen LeCraft Henderson dissenting. The department designated Anthropic a supply chain risk in March, and the designation stays in place.
Jury finds Facebook liable for deceiving users in Cambridge Analytica case A New Mexico jury found Facebook liable on Friday for deceiving users about privacy protections, after a two-week trial in Santa Fe tied to the Cambridge Analytica harvest of roughly 87 million profiles. Jurors found the failure affected the state’s entire population of more than 2 million people. The state is seeking the maximum $5,000 per violation, and the judge will set the total.
Unsealed filings in Authors Guild v. OpenAI Newly unsealed briefs in the Authors Guild’s lawsuit against OpenAI and Microsoft over training on copyrighted books quote OpenAI’s Sam McCandlish worrying about the optics of “openai uses copyrighted data from sketchy russian website” showing up on Hacker News. The Guild says the filings show executives at both companies knew the conduct was unlawful. These are the plaintiffs’ briefs, not court findings, and the case is still in litigation.
Tech Industry
ASML says it sold ‘absolutely nothing’ in Europe in 2026 ASML, the only supplier of EUV lithography systems and Europe’s most valuable company at around $660 billion, reported 0% of its revenue from Europe in the first half of 2026, down from 1% in 2025 and 5% in 2024. Frank Heemskerk, executive vice president of public affairs, said on Dutch television that the company is “selling absolutely nothing in Europe” because Europe is not investing and no chip factories are being built there. ASML is asking EU governments to help guarantee demand for European-made chips.
Economy & Labor
California is chasing wealth that has feet California has certified a Billionaire Wealth Tax for the November ballot: a one-time 5% levy on the state’s billionaires, paid over five years, to raise about $20 billion a year for health care, food aid and schools after federal cuts. A Center for Land Economics report argues the tax will fall short because billionaires can move and land cannot, and estimates the state’s total land value at about $8.14 trillion.
That’s what I’m watching. What caught your attention this week?
-Eric