Open source AI grows, but companies still spend most on Anthropic

In short: Cheaper open source AI tools are being used more, but many businesses are still spending most of their AI budget on premium models like Anthropic.

What's going on

A new argument from Decagon CEO Jesse Zhang says open source AI and top paid AI models are not always direct rivals. Open source here means the AI model can be downloaded and used more freely, like getting a recipe instead of ordering the meal. Zhang says companies often use the most advanced and expensive models first to test new ideas, then later switch to cheaper models once the work becomes routine.

Some recent data points line up with that story. Vercel, a company that tracks AI use through its “AI gateway” (a kind of traffic controller for AI requests), shows DeepSeek leading by total activity in the past week, measured by “tokens.” Tokens are small chunks of text that AI reads and writes, like counting how many words the AI processes.

But on Vercel’s dashboard, Anthropic still accounts for more than half of total spending on the platform, even though its share of raw activity is lower. Another tracking site, OpenRouter, shows a similar pattern. DeepSeek V4 Flash leads in usage, but Anthropic’s Opus 4.8 is far more expensive per token, which likely means it still brings in a large share of revenue.

What to watch

If Zhang is right, “premium” AI may keep winning the early stage work where accuracy matters most, while open source models take over day to day production work. The big question is whether cheaper models will become good enough to replace premium models for harder tasks, or whether a two tier market will stick around.

Source: TechCrunch AI