Quick Hits — September 4, 2026
A quick catch-up on the day's mid-tier AI stories — benchmark arguments, app-store politics, and a Chinese-lab capital plan that just broke cover.
ARC-AGI-3 reshapes the AGI timeline as Chollet concedes the pace is "twice as fast" as expected. ARC Prize ran GPT-6 Astra against its third-gen reasoning suite and, with its own harness, found the model cleared 96% of levels in fewer moves than the median human solver — about half the moves, on average. On a cost-per-game basis Astra came in roughly twice as efficient as the average human tester ($12.78 vs. Astra's fraction-of-a-cent equivalent). François Chollet used the results to publicly revise his AGI forecast: "Sooner, because progress is happening faster than I expected." ARC Prize is already building ARC-AGI-4 with a Q1 2027 release target.
Chinese app stores are reportedly floating a "double-authorization" rule that could lock AI agents out of mobile apps. Sina reports that major Chinese app stores are drafting a dual-consent requirement — the AI agent and the host app would both have to approve every interaction before an agent could read or tap a screen. It's the agent equivalent of Apple's App Tracking Transparency, and critics argue it would create a walled garden that only the largest platforms could negotiate through. Smaller developers and Western agents would effectively be cut off from China's mobile surfaces.
Moonshot is preparing a Hong Kong IPO that could raise $3B–$5B as soon as this year, Bloomberg sources say. The Kimi maker is weighing a listing on the Hong Kong exchange, with a target raise in the $3 billion to $5 billion range and a window that opens in the back half of 2026. The move would extend China's 2026 AI capital wave beyond Zhipu and DeepSeek and put Moonshot on a path toward the same "tier-one" funding tier as its domestic peers — though sources cautioned the size and timing are still preliminary.
Sources: The Decoder · Sina · Techmeme
