Following·Day 2·first covered 22 Sep 2026·6 sources·updated 22 Sep 2026

Chinese open models' lead over US labs

The gist

Open-weight models at 78% of token volume; closed models hold revenue lead

The split between token volume and revenue reflects two distinct competitive dynamics: open-weight models compete on price for commodity workloads while closed models retain pricing power on higher-value tasks. MilkRoadAI warned that if open models keep closing the performance gap, closed labs could lose that pricing advantage.

The full picture

Open-weight models account for 78.4% of token volume on Vercel's AI Gateway, up from roughly 40% in June, driven largely by DeepSeek V4.1 Flash at 59.3% of tokens. Closed-model providers retain the bulk of revenue: Anthropic's five models capture 36.9% of total spending while handling a small fraction of tokens, and Claude Opus 4.8 alone accounts for 13.7% of spending on 1.7% of token volume. MindStudio, tracking Vercel data from June to August, identified the crossover point where open-weight models surpassed closed-weight models in total token share, with DeepSeek passing Anthropic at 25.2% versus 24.5%. Open-weight models can run at roughly one-tenth the price of closed models, making them attractive for non-premium enterprise use cases. At the company level, OpenAI stands at approximately $40 billion in annual recurring revenue and Anthropic at approximately $65 billion, while all major Chinese AI developers combined generate roughly $10.7 billion, with DeepSeek at an estimated $500 million.

How it developed

22 September 2026

Epoch AI's JS Denain and Nathan Lambert debate distillation's role in Chinese labs' capability gains; Denain updates toward distillation being a larger factor than previously thought.

A briefing for U.S. Congressional members, published September 22, finds Chinese open-weight models score 42-45 on the Artificial Analysis Intelligence Index against 23-26 for American open models, hold a 1.6 billion Hugging Face download lead, and appear in over 40% of AI/ML arXiv papers against roughly 30% for U.S. models; distillation of American frontier models accounts for only an estimated 1-2 months of the gap. Vercel AI Gateway data shows open-weight models at 78.4% of token volume, with DeepSeek V4.1 Flash at 59.3% of tokens but only 5.1% of spending.

First citedInterconnects Milk Road AI Twitter Rohan Paul Twitter

21 September 2026

Congressional briefing published analyzing Chinese lead in open-weight model capabilities, downloads, and academic citations.

20 September 2026

Alibaba releases Qwen-Image-2.1, a 7B open-weight image generation and editing model with ComfyUI and vLLM-Omni support.

19 September 2026

Open-weight models at 78.4% of Vercel AI Gateway token volume; DeepSeek V4.1 Flash at 59.3% of tokens but 5.1% of spending; Anthropic's five models at 36.9% of spending

27 August 2026

MindStudio identifies open-weight usage crossing above closed-weight in June-August Vercel data; DeepSeek surpasses Anthropic in token share at 25.2% vs 24.5%

23 August 2026

Open-weight models reach 62% of Vercel AI Gateway traffic; Anthropic's closed models at 61-65% of spending

Sources

Qwen Twitter

Rohan Paul Twitter

Interconnects

Milk Road AI Twitter