AI News for July 20, 2026 — 7min.ai

[NEWS] Alibaba previews Qwen 3.8, its most powerful model yet, calls it second only to Fable 5: Alibaba unveiled Qwen3.8-Max-Preview, a 2.4-trillion-parameter model that trails only Anthropic's Claude Fable 5 among frontier systems.
[RESEARCH] Claude Fable produces a counterexample disproving the Jacobian Conjecture: Anthropic's Claude Fable produced an explicit counterexample to the Jacobian Conjecture, a decades-old open problem in algebraic geometry.
[RESEARCH] Google DeepMind says video generators already hold the world models vision has lacked: Google DeepMind introduced GenCeption, a model that repurposes a pre-trained video generator for computer vision tasks like depth estimation, segmentation, and 3D pose estimation.
[NEWS] SpaceX quietly builds an AI compute business renting GPUs to Anthropic and Google: SpaceX rents unused Colossus data center capacity to AI leaders, agreeing in May to give Anthropic access to roughly 325,000 Nvidia GPUs for $1.25 billion per month and Google roughly 110,000 GPUs at $920 million per month.
[NEWS] Moonshot pauses new Kimi subscriptions as K3 demand strains its GPUs: Moonshot AI temporarily halted new Kimi subscriptions after K3 drew far more demand than expected, pushing compute close to current capacity limits over 48 hours.
[OPINION] Kimi K3 reignites Silicon Valley's open vs closed AI war: Moonshot's Kimi K3 has split American AI leaders over open-weight strategy, with a weekend debate erupting on X.
[RESEARCH] Kimi K3 tops Fable 5 in frontend coding but trails badly on hard math: Benchmark data shows Kimi K3 performing strongly on practical coding but struggling with complex reasoning.
[NEWS] Kai-Fu Lee's 01.ai targets a 2027 Hong Kong IPO: 01.ai, the Beijing AI startup founded by computer scientist Kai-Fu Lee, is raising a pre-IPO round ahead of a planned Hong Kong stock exchange listing in 2027.
[RESEARCH] Study: AI advice made people 3x less accurate but twice as confident: Researchers from French and Italian universities found that AI advice collapsed willingness to say "I don't know" from 44% to 3%, while accuracy dropped from 27% to 9% and confidence rose from 30% to 76%.
[TOOLS] Apple pilots 'Live Notes' AI to record and transcribe Genius Bar sessions: Apple is testing Live Notes at some retail stores to automatically record and transcribe Genius Bar appointments.

About 7min.ai · Subscribe to newsletter