While US markets closed the week higher (Dow +478 points, Nasdaq +0.48%), the AI industry just had its biggest news week of 2026. Everything below is verified against Hacker News discussion data (Sept 18–26).
1. Claude Opus 5.5 vs GPT-6: the same-week showdown
Anthropic shipped Claude Opus 5.5 and OpenAI shipped GPT-6 (Sol and Luna variants) within 48 hours of each other. The two launch threads pulled a combined 3,569 points and nearly 2,000 comments on Hacker News — numbers usually reserved for major CVEs, not product launches.
The Claude camp praises long-horizon agentic coding. The GPT-6 camp praises multimodal breadth. The sharpest recurring complaint about GPT-6: it got nicer and slightly less honest.
2. AI starts producing science, not just text
Two separate results landed the same week. Anthropic announced Claude identified a previously unknown enzyme system with CRISPR-like repeats. Meanwhile, someone used GPT-6's Astra model to crack an Enigma ciphertext that had been unresolved since 2005.
AI moving from summarizing knowledge to generating it has direct pricing consequences for every experience-based profession.
3. A Microsoft exec calls AI scraping 'the biggest labor theft in history'
That quote, from a Microsoft executive — the largest corporate backer of OpenAI — drew 954 points and 839 comments. When the money side of the industry starts questioning training-data legitimacy in public, copyright litigation is coming.
4. Washington stops treating AI as just business
A US appeals court upheld the designation of Anthropic as a government supply-chain risk. The same week, reporting revealed the NSA is spending billions testing frontier models. Strategic-asset treatment is now official.
5. China open-sources a medical AI covering 150 diseases
Alibaba released an open-weight medical model able to help detect cancer and roughly 150 conditions (SCMP). Different playbook from the US labs: no ticket price, distribution first.
6. The one that touches your wallet: FT tested AI on money questions
The Financial Times ran money questions through chatbots and found they get most answers wrong. Consistent with what we see building daily sentiment indices from 10,000+ retail comments: these models narrate beautifully and hallucinate numbers confidently.
Also this week
- Mercury 2.5 hit 770 tokens/second (Artificial Analysis)
- OpenAI was reported to employ a 'net army' of reputation workers
Yijinmoming Finance publishes a daily China retail-investor sentiment index built from 10,000+ raw comments scraped across 8 platforms (East Money, Sina, Xueqiu, Weibo and more). Latest reading: 62.5/100, up 8.2 from the holiday start. Full recap with sources: Telegra.
Top comments (0)