Skip to content
Source · Daily Brief

AI Daily Brief — 29 November 2025

Saturday belonged to DeepSeek again. DeepSeek released Math-V2 on Hugging Face + GitHub under permissive license — first open-weight model to reach IMO gold-medal tier. Solved 5 of 6 IMO 2025 problems (matched by only ~8% of human contestants). Scored 118/120 on Putnam 2024 — surpassing top human score of 90. Previously only Google DeepMind + OpenAI proprietary systems had reached this bar. Math-V2 uses self-verification — model checks own derivations step by step before committing; key reason it generalized beyond IMO to Putnam + Chinese Mathematical Olympiad. Released with weights, making approach reproducible by open community. Post-Thanksgiving US frontier-lab quiet — no major OpenAI/Anthropic/Google DeepMind/Meta releases Saturday itself. Week prior saw Opus 4.5 + Gemini 3 land back-to-back. AI shopping agents drove record Black Friday weekend traffic: AI-referred traffic to US retail up ~500-700% YoY heading into Sat-Sun. Target enabled full in-ChatGPT checkout; Best Buy dominated AI “best deal” queries across computers, appliances, phones, TVs. Cyber Monday queued for record $14.25B online. AI funding momentum carries into weekend — November closed as largest mega-round month in 3 years: 43% of all venture funding to 14 companies raising $500M+ each. Headline deals: Cursor $2.3B at $29.3B (NVIDIA + Google strategic), Bezos Project Prometheus $6.2B, Lambda $1.5B (TWG Global lead). AI Fraud Deterrence Act circulating in Congress over Thanksgiving recess — Rep. Ted Lieu (D-CA) + Rep. Neal Dunn (R-FL) bipartisan bill from earlier November. Up to 30 years + $2M for AI-assisted bank fraud; 20 years + $1M wire/mail fraud. Adds AI-mediated deception to existing fraud statutes with explicit First Amendment carve-outs for satire + parody.

Top stories

  • DeepSeek Math-V2 — first open model with IMO gold. 5/6 IMO 2025 problems. 118/120 Putnam 2024 (top human 90). Hugging Face + GitHub permissive license. via SCMP
  • Math-V2 self-verifying reasoning. Model checks own derivations step by step before committing. Generalized beyond IMO to Putnam + Chinese Mathematical Olympiad. via OfficeChai
  • Post-Thanksgiving frontier-lab quiet. No US lab releases Saturday. Week prior had Opus 4.5 + Gemini 3 back-to-back. via Aishwarya Srinivasan
  • AI shopping +500-700% YoY into weekend. Target full in-ChatGPT checkout; Best Buy dominated AI “best deal” queries. Cyber Monday queued for $14.25B record. via Axios
  • November largest mega-round month in 3 years. 43% of VC to 14 companies raising $500M+. Cursor $2.3B/$29.3B + Bezos Prometheus $6.2B + Lambda $1.5B (TWG Global). via Crunchbase
  • AI Fraud Deterrence Act circulating in Congress. Lieu + Dunn bipartisan. 30 years + $2M for AI bank fraud; 20 years + $1M wire/mail. First Amendment satire/parody carve-out. via Rep. Ted Lieu

Who shipped

DeepSeek shipped Math-V2. OpenAI, Anthropic, Google, Meta, xAI, Alibaba all dark on launches.

Open-source pulse

Math-V2 reshapes the math-reasoning open-weight landscape overnight. Combination of permissive license + IMO gold + self-verification is unprecedented. China-led open-weight surge continues — Math-V2 + Qwen + Hunyuan-1.5 + Olmo 3 trifecta defining November’s release wave.

Money, infra & hardware

November’s mega-round momentum into Cyber Monday weekend. $11.8B Black Friday US online + 5-7× AI-shopping traffic uplift framing the Cyber Monday cycle as agentic-commerce’s first scaled holiday.

Quiet corners

AI Fraud Deterrence Act is the substantive US-policy moment — bipartisan US House push to recognize AI-mediated deception as separate fraud category with severe penalties.

By the numbers

  • 5/6 / 118/120 / 90 — Math-V2 IMO 2025 problems solved / Putnam 2024 score / prior top human
  • +500-700% YoY — AI-referred retail traffic weekend
  • $14.25B — Cyber Monday queued record
  • 43% / 14 / $500M+ — November VC concentration / mega-round count / per-round threshold
  • 30 years / $2M — AI Fraud Deterrence Act max penalties (bank fraud)
  • Most-mentioned company: DeepSeek
  • Quietest segment: US frontier-lab launches

Compiled by AI Feed’s editor from verified web sources for 29 November 2025.