US China AI Gap

The US lead in AI over China has narrowed to 3% on benchmark scores, a record low.

Bloomberg Intelligence says the gap was about 15% earlier this year and about 9% in May. The turning point was DeepSeek's V4.1 Flash, released in September, which now ranks sixth globally on LiveBench with a score of 81.1. Anthropic's top score on the same board is 83.4.

Here is my point of view. The leaderboard is the wrong thing to watch. When a model that trails the frontier by a few points can be built under chip restrictions, model quality stops being a durable moat. The lasting advantage moves to proprietary data, the workflows wired into your business, and the ability to switch models when the ranking changes.

Any company that builds its AI strategy around one vendor's model is betting on something that gets more fragile every quarter.

Which part of your AI stack would you protect first if the frontier gap keeps closing?