AI Market in Flux: Companies Are Switching Models More Often

Open models are taking over

In September 2026, open-weight models handled 71.5 percent of all tokens (small chunks of text that AI reads and writes) sent through Vercel's AI Gateway. On OpenRouter, another AI routing service, they passed 70 percent for the first time in June. By spending, they reached 23 percent of estimated costs on AI Gateway.

Key numbers

• Open-weight share of tokens on AI Gateway in September 2026: 71.5 percent
• Open-weight share of spending in June: about 5 percent; in September: 23 percent
• Spending on open-weight models grew more than 70 times since December 2025
• Spending on proprietary frontier models (top closed models) grew about 10 times in the same period
• Anthropic's share of Gateway spending fell from 63.7 percent to 43.9 percent in one month

Why models are switching

Teams now treat AI models almost like swappable parts. Teams with at least one billion tokens used an average of 42 different models. Vercel sees this as a sign that models are becoming interchangeable. Note that these numbers come only from Vercel's own traffic and use estimated list prices, not real invoices.

Why it matters

Companies can save money or choose the best tool for each job by switching models. But China's proposed export rules add risk. The draft, published on October 16, covers 23 AI technologies, including large models with at least two trillion parameters (the numbers that set how a model learns). Experts warn that Chinese authorities could limit license deals at any time. This would affect US companies that host or sell Chinese models.

What to watch

• The public comment period on China's draft export rules runs until November 15, 2026
• Whether China limits license deals, even though API access and already published open-source tech are meant to stay allowed
• How US companies that host Chinese models respond