Open models are now capturing 30% of tokens, tripling in one year

Signals Inbox·July 27, 2026·Horizontal AI

Open models have jumped from roughly 10% of AI tokens to 30% in one year. The bigger signal is not just adoption. On Vercel’s AI Gateway, open-weight models recently handled 29% of tokens while generating less than 4% of spending. Developers are not only testing open models. They are using them to buy far more intelligence for the same money.

Q1What actually happened?

Together AI said open models now account for around 30% of AI tokens, up from about 10% one year earlier. A separate Vercel measurement found a very similar number: open-weight models handled 29% of its AI Gateway tokens in June 2026, up from 11% in April.

Q2What does 30% of tokens mean?

Tokens are the pieces of text that AI models read and generate. More tokens usually mean more actual model usage. So this is stronger than a download count, a benchmark score, or developers saying they like open source. It suggests open models are handling a meaningful share of live AI workloads.

Q3Why are open models gaining so quickly?

Cost is the clearest answer. Vercel found that open models processed 29% of tokens but represented less than 4% of spending. Together AI says some customers save between 6 and 60 times compared with closed-model pricing. When an application processes billions of tokens, even a small price gap becomes a major bill.

Q4Are open models now better than closed models?

Not across every task. Closed models from OpenAI, Anthropic, and Google can still lead on difficult reasoning, coding, and reliability tests. But many production requests do not need the absolute best model. They need a model that is fast, controllable, and good enough at a much lower price. That is where open models can take huge volume.

Q5Who is gaining from this shift?

Open-model makers such as DeepSeek, Qwen, Kimi, MiniMax, and Meta gain distribution. Infrastructure companies such as Together AI, OpenRouter, and other inference providers also benefit because they let developers switch between models without rebuilding their applications. The model becomes more interchangeable, while the routing and infrastructure layer becomes more valuable.

Q6Why does this threaten closed-model companies?

Closed labs spend billions building frontier models and need premium prices to recover that cost. If developers send routine work to much cheaper open models, closed providers could keep the hardest tasks but lose a large amount of everyday volume. That creates pressure to lower prices, release smaller models, or offer open weights of their own.

Q7So what should we watch next?

Watch whether the 30% share keeps rising as open models improve, and whether that usage moves deeper into large companies. Also watch spending, not only tokens. Open models already generate lots of volume at very low prices. The real power shift arrives if they keep that usage while building a profitable business around it.