A year ago, the consensus was clear: open source AI would commoditize the model layer and destroy companies like Anthropic. Llama, Mistral, Qwen — the flood of free models would make paying for Claude pointless.
It has not happened. Anthropic is reportedly growing faster than ever. Claude usage is surging. Enterprise adoption is accelerating.
Why?
The answer reveals something uncomfortable about the AI market that most indie developers do not want to hear.
Models are not the product
The open source community made a critical error: they assumed AI was a model problem. Build a better model, win the market.
But Anthropic is not selling a model. They are selling a reliable, maintained, supported AI service. The model is just one component.
When you use Claude, you get:
- Consistent uptime and performance
- Regular model improvements without migration pain
- Enterprise support and compliance certifications
- API stability and backward compatibility
- Safety features and content filtering
When you use an open source model, you get:
- A model file
- Good luck
The indie developer trap
I see this mistake constantly in indie hacker communities. Someone discovers Llama 3, runs it locally, gets impressive results, and declares they will never pay for an API again.
Three months later, they are back to Claude or GPT. The reason: maintaining your own AI infrastructure is a full-time job.
- Model updates break your prompts
- Performance tuning never ends
- Scaling is your problem
- When something breaks at 2am, you are the support team
This is the same trap that made SaaS win over self-hosted software. Not because SaaS is better technology, but because managing infrastructure is not most people's core competency.
What this means for builders
If you are building AI-powered products, the choice is not binary. Smart teams use both:
- Claude/GPT for production — reliability, support, compliance
- Open source for experimentation — customization, cost control, learning
The key is having infrastructure that supports both. This is where MonkeyCode gets interesting. It is model-agnostic by design. You can connect Claude for your production workflows and use open source models for experimentation. When an open source model matures enough for production, you switch without rebuilding your entire stack.
That flexibility matters more than any single model's benchmark score.
The real competition
Anthropic's real competition is not open source models. It is other API providers — OpenAI, Google, Cohere — who offer the same reliability and support.
Open source AI is valuable. But it serves a different market: researchers, experimenters, and teams with the engineering capacity to run their own infrastructure.
For everyone else, paying for a reliable API is the rational choice. And that is why Anthropic is not dying.
Running your own models or paying for APIs? What is your setup? Always curious how other indie hackers approach this.
Top comments (0)