AI Weekly | 2026-W37

2026-09-13 · AI Weekly

weeklyai

AI Weekly | 2026-W37

Week 2026-W37 (Sep 7–13) — From OpenAI’s claimed solution to a Millennium Prize Problem to a record-breaking week for open-source model releases, this edition covers the moments that defined AI this week.

Weekly Top 5

No curated top articles were available in the pool this week. This section will be restored when scoring data resumes.

AI Highlights

  • OpenAI shared a solution to the Navier-Stokes Millennium Prize Problem, claiming a proof completed by agents using its next-generation model (stronger than GPT-6 Astra). The post drew ~103K points on X, making it the week’s most-discussed AI event, though OpenAI later clarified its proof differs from two mathematicians’ independent work and that it did not access their specific data.
  • Anthropic announced Claude completed the first formal proof of Fermat’s Last Theorem, described as the largest Lean proof to date — a milestone for AI-assisted formal mathematics verification.
  • Anthropic’s economics team released a 2030 AI economic impact model, simulating potential effects on growth, employment, and wages across scenarios.
  • Asahi Linux officially supports Apple M3 Macs, extending its Linux-on-Apple-Silicon support with noted limitations.
  • Anthropic restricted Claude to users 18 and older, updating terms of service across all products; the change sparked developer debate about AI access in education for minors.

New Releases

  • Inference.net released Schematron V2 Turbo (2026-09-12), a turbo-tier addition to its Schematron inference series.
  • Inference.net released Schematron V2 Small (2026-09-12), a smaller, lighter variant for cost-sensitive deployments.
  • Sakana released Fugu Ultra v2 (2026-09-11), the upgraded flagship of its Fugu model line.
  • Sakana released Fugu Max (2026-09-11), a new high-capacity model in the Fugu series.
  • inclusionAI released Ling 3.0 Flash VL (2026-09-10), a vision-language model iteration in the Ling series.

Usage Rankings

  • Tencent: Hy4 preview climbed to rank 1 (up 14 positions from 15) — the week’s biggest mover into the top spot.
  • Z.ai: GLM 5.3 Flash rose to rank 3 (up 17 from 20), continuing strong momentum.
  • DeepSeek: V4 Flash 0731 jumped to rank 4 (up 45 from 49), with V4 Flash 0423 also surging to rank 6 (up 143 from 149).
  • Xiaomi: MiMo-V2.5 entered the top 5 at rank 5 (up 147 from 152), the largest positional gain this week.
  • Falling: no major decliners were reported in the top tier this week.

Trend Summary — The week was defined by two parallel narratives: frontier labs pushing the boundaries of AI-assisted formal mathematics (OpenAI’s Navier-Stokes claim, Anthropic’s Fermat proof), and a surge of open-source model releases on ModelScope (Qwen3.8-27B, Kimi-K3, GLM-5.3) alongside strong popularity gains for Chinese-origin models in usage rankings.