AOYii AI Daily - Nov 22, 2026

Highlights

  • Anthropic mathematician claims Riemann proof with Claude (Reuters)
  • Trump: US-China agree military AI needs human control (Reuters)
  • Anthropic falls 10% day after record IPO (Epoch Times)
  • Zhipu to launch 3T-parameter flagship model in Q4 (Bilibili AI Daily)
  • Siemens and Agibot unveil agent AI robot gripper workflow (Formnext)

Headlines

[Anthropic mathematician Levent Alpoge claims a proof of the Riemann Hypothesis, working with Claude] On November 20, Levent Alpoge, a mathematician on Anthropic's technical staff, posted a four-page note on his personal website: "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." Posed by Bernhard Riemann in 1859, the conjecture concerns the distribution of prime numbers and is widely considered the most important unsolved problem in pure mathematics, unproven for 167 years. Alpoge says the method extends to the Generalized Riemann Hypothesis and includes a 20-step timeline of the collaboration with Claude; the claim has not been peer-reviewed and mathematicians urge caution. (Source: Straits Times/MLQ)

[Trump: US and China agreed military AI decisions must keep humans in the loop] On November 20, US Treasury Secretary Scott Bessent told CNBC that following the Trump-Xi summit in Shenzhen, the two leaders agreed that in military decisions using AI "humans should always be involved" and that AI should never autonomously make major military actions. Trump added the two sides would set up "some joint guardrails" on AI, and warned he would penalize US companies that build data centers abroad, saying AI compute must stay in America. It is the most concrete AI-governance outcome of Trump's Asia trip. (Source: Reuters/Bloomberg/LA Times)

[Anthropic sinks 10% in second day of trading as AI sentiment stays fragile] On November 20 in New York, Anthropic fell as much as 14% pre-market after its record IPO, with shares nearing the $83 offer price before trading around $87.37, down about 10%, mid-morning. The strong 17% debut did not lift the broader tech sector: the Nasdaq and Dow kept sliding and Asian equities hit seven-month lows, as investor concerns over AI valuations and circular financing continued to build. (Source: Epoch Times/WSJ/Investing.com)

Products & Tech

[Zhipu teases a 3-trillion-parameter flagship model for Q4] On November 21, Zhipu disclosed plans to release a 3-trillion-parameter flagship model in the fourth quarter, continuing its open-plus-closed dual-track strategy. The company earlier shipped the open-source GLM-5.2, built for long-horizon coding and agentic workflows with a 1-million-token context window, positioned against Claude Opus 4.8; in Hong Kong trading, Zhipu and MiniMax recently fell as much as 11% each in the wake of Anthropic's IPO. (Source: Bilibili AI Daily/Commercial Times/Anue)

[Xiaomi launches V2.6 Pro high-speed version of its large model] On November 21, Xiaomi rolled out the high-speed V2.6 Pro version of its large model, targeting lower inference latency and faster on-device response for high-concurrency scenarios such as smartphones and in-car systems, the latest move by a Chinese vendor in the inference-efficiency race. (Source: Bilibili AI Daily)

[Siemens and Agibot launch agentic AI that designs 3D-printed robot grippers] On November 21 at Formnext 2026 in Frankfurt, Siemens, Agibot, EOS and AM Solutions launched an agentic AI workflow that designs 3D-printable humanoid robot grippers from text prompts, with the system orchestrating engineering tools to shorten hardware iteration cycles. (Source: emergent.sh)

IPO & Fundings

[Anthropic closes out IPO and private placement: $107.3B raised, a global record] Anthropic's IPO settled on November 20: the public offering priced at $83 a share, raising about $81.3 billion, plus a $26 billion private placement to Alphabet, Microsoft, Nvidia, AMD and Broadcom, for a combined $107.3 billion — the largest IPO in history, surpassing SpaceX's record, at a fully diluted valuation of about $1.5 trillion. (Source: Anthropic/Business Insider)

[Zhipu's Hong Kong IPO nears as China model pricing faces a reset] Sina Finance analysis argues that after Anthropic's trillion-dollar IPO, unlisted model companies face a valuation reset — from "do you have compute" to "is the compute worth it," with revenue growth, customer mix, inference cost and cash flow becoming the core pricing factors. Zhipu and MiniMax have both filed for Hong Kong listings, with hot retail demand and listing-day risk coexisting. (Source: Sina Finance/36Kr)

Open Source & Research

[Karpo open-sources Decision 31B: native multimodal decision model tops Decision Index] Karpo released Decision 31B, a decision model built on Gemma 4 31B, with all weights and inference code open, plus smaller 0.8B-to-12B siblings. Unlike generative models, it outputs per-option probabilities in a single forward pass, suited to high-frequency judgment tasks such as triage, ticket handling and safety review; the vendor claims a 73.9 score atop the Decision Index of 300 models with a 95ms median latency on one RTX PRO 6000 (results are largely vendor-reported). (Source: Zhihu AI Briefing)

[Tianhe model V2 launches: China's South-to-North Water Diversion upgrades its sector model] On November 21, China South-to-North Water Diversion Group released the upgraded Tianhe V2 model, deepening sector deployment in water dispatching and infrastructure monitoring — the latest iteration of state-enterprise industry models. (Source: Capital Futures)

Company & People

[Altman confirms OpenAI will not go public in 2026] OpenAI CEO Sam Altman confirmed on Saturday that the company will not go public in 2026, saying concerns around AI safety make this an ill-advised time for a listing; OpenAI recently raised the first tranche of its new round — about $11 billion at a $1.25 trillion valuation, below the $1.4 trillion originally sought. (Source: The Business Standard/Bloomberg)

[DeepSeek tops $1B annualized revenue after raising 100B yuan toward an IPO] DeepSeek raised about 100 billion yuan (roughly $15 billion) in its latest round, with CATL and Tencent among the largest investors; The Information reports its annualized revenue run-rate has passed $1 billion, with funds going to model training, compute buildout and a planned STAR Market IPO. (Source: Qianzhan/The Information)

Policies & Data

[Trump warns of penalties for US companies building data centers abroad] On November 20, after returning from his Asia trip, Trump said he would penalize US companies that build data centers overseas, insisting AI compute infrastructure must remain on US soil; the statement came alongside the US-China AI guardrails consensus and could reshape hyperscalers' expansion plans. (Source: LA Times/Al Jazeera)


Powered by AOYII.com

Related News