Skip to content
Source · Daily Brief

AI Daily Brief — 9 November 2025

Sunday belonged to Kimi aftermath three days post-launch. Consensus framing crystallized: “closest open weights have come to closed frontier” — analogous to DeepSeek R1’s fast-follow to o1. Top scores on Humanity’s Last Exam + BrowseComp; 200-300 sequential tool calls without human intervention. Simon Willison’s hands-on notes: K2 Thinking hit 93% on tau-squared-Bench Telecom — highest agentic tool-use score Willison has personally measured; produced realistic multi-agent task graphs with dependency reasoning via OpenRouter. Weekend commentary placed K2 Thinking alongside DeepSeek + Qwen as the three Chinese open-weight families that became household names by end of 2025 — Chinese models on track to take 7 of top 10 most-downloaded slots on Hugging Face by November/December. Wall Street starts losing faith in AI hype (TechCrunch Nov 8 surfaced widely Sunday): investor skepticism rising heading into the BofA fund-manager survey window (Nov 7-13) that would later show net 20% of managers calling companies over-invested for the first time since 2005. OpenAI-AWS $38B compute deal + Apple-Google ~$1B/year Gemini-Siri still anchoring the Nov 3-9 week. Google Ironwood TPU (7th-gen) hardware talking point — 4× perf over prior gen, hyperscaler answer to Blackwell.

Top stories

  • K2 Thinking aftermath: open weights closing on closed frontier. Three days post-launch consensus — analogous to DeepSeek R1’s fast-follow to o1. Top HLE + BrowseComp; 200-300 sequential tool calls. via Interconnects
  • Willison: K2 Thinking 93% on tau-squared-Bench Telecom. Highest agentic tool-use score Willison has personally measured. Produced realistic multi-agent task graphs with dependency reasoning via OpenRouter. via Simon Willison
  • Chinese open-weight surge: 7 of top 10 HF slots. K2 Thinking + DeepSeek + Qwen on track for 7 of top 10 most-downloaded slots on Hugging Face by Nov/Dec. via StartupHub.ai
  • Wall Street faith in AI hype falling. Investor skepticism rising heading into BofA survey window (Nov 7-13) that would show net 20% of managers calling companies over-invested for first time since 2005. via TechCrunch
  • OpenAI-AWS $38B + Apple-Google Siri deals still anchor week. Two structural infra commitments framed the Nov 3-9 narrative. via LinkedIn recap
  • Google Ironwood TPU 7th-gen. 4× perf over prior gen, hyperscaler answer to Blackwell roll-out. via Google Blog
  • Yang Zhilin Reddit AMA scheduled Tue. Moonshot founder + partners to take questions on K2 Thinking — expected to address $4.6M training-cost claim. via 36Kr

Who shipped

Nobody Sunday. OpenAI, Anthropic, Google, Meta, xAI, DeepSeek, Alibaba all dark on launches.

Open-source pulse

K2 Thinking continues to dominate — viral analyses and bench scores driving “DeepSeek moment” framing into Tuesday AMA.

Money, infra & hardware

No new rounds Sunday. Bubble fears from TechCrunch Saturday piece keep building heading into BofA fund-manager survey window.

Quiet corners

Sunday traffic was dominated by analysis pieces digesting prior week (K2 Thinking, OpenAI-AWS, Apple-Google Siri) plus anticipation of Yang’s AMA.

By the numbers

  • 0 — frontier-lab launches Sunday
  • 3 days — since K2 Thinking landed, still dominating analyses
  • 93% — K2 Thinking on tau-squared-Bench Telecom (Willison)
  • 7 of 10 — projected Chinese share of HF top trending by year-end
  • $38B / ~$1B/yr — OpenAI-AWS / Apple-Google Siri pact still anchoring the week
  • Most-mentioned company: Moonshot AI
  • Quietest segment: everything except K2 Thinking analysis

Compiled by AI Feed’s editor from verified web sources for 9 November 2025.