The Open-Source Tipping Point Is Closer Than You Think

Six months ago, the top 20 was wall-to-wall proprietary. Claude, GPT, Gemini, and a bunch of models with corporate logos.

This week there are three open-source models in the top 20. Mistral, DeepSeek, and Qwen. All three are climbing.

What changed

Not one thing. Four things, happening simultaneously:

Price. Open-source models are 60-80% cheaper per token than proprietary alternatives. For high-volume use cases, that math is hard to ignore.

Quality. The gap between open and closed models has shrunk dramatically. On coding tasks specifically, Mistral’s latest is within striking distance of GPT-4.

Control. You can fine-tune an open model. You can host it yourself. You can inspect the weights. Enterprise compliance teams love this.

Speed. Community optimization of open models — quantization, distillation, specialized inference engines — often makes them faster than the original versions.

The chart that explains it

I looked at the weekly growth rate of the top 5 open-source models versus the top 5 closed-source models over the past 8 weeks.

Open-source average weekly growth: +4.2% Closed-source average weekly growth: +1.1%

The open models are growing nearly 4x faster. They’re starting from a smaller base, so the absolute numbers still favor proprietary. But trajectories diverge this clearly, you pay attention.

What this doesn’t mean

Open-source isn’t about to take the #1 spot. Not this quarter, probably not this year. Claude’s lead is real and the quality gap at the very top is still meaningful.

What it means is: for the 80% of use cases that don’t need state-of-the-art, open-source is becoming the rational economic choice.

What I’d do

If you haven’t tested an open-source model for your workload in the past month, test one this week. Mistral if you’re doing code. Qwen if you need vision. DeepSeek for general-purpose.

You might be surprised.