This might surprise many, but the state of OpenAI large language models are not far enough ahead of their competitors to be a serious barrier to competition. I use models from OpenAI, Anthropic, and Google every day. When any one of them goes down, I can swap over to another competitor pretty easily¹. This doesn’t mean they don’t have unique strengths, just that none are so much stronger than the others that I can’t swap them in and out without missing a beat.

The differences that exist are primarily in the frontier models—the top models from each competitor². However, as soon as you step down from that top tier, there’s really no significant difference in the mid-tier and low-tier models. For the majority of use cases, one of those lower or mid-tier models will be sufficient.

That includes the open-source models Gemma, Llama and Deepseek.

Where the real differences lie is in the systems each vendor is building around their core models.

Anthropic’s best interface is their coding tool, followed perhaps by Claude Desktop, but both are feared torwards technical people.

Google’s interface is minimal, focusing on providing access to some of the most advanced models without a lot of extra features.

OpenAI, on the other hand, has created a web interface with significant differentiation through

• Integrations with third-party platforms like Box, Dropbox, Microsoft, and Google³
• Canvas
• Infinite-length conversations
• Advanced voice mode⁴
• Memories
• The ability to do analysis in languages like Python

These are NOT part of the model, they are all part of the software interface they’ve built on top of the models. The implications here are that OpenAI’s strategy is about locking users into ChatGPT, not just the underlying models.

This is why OpenAI is allegedly considering their own productivity suite like Microsoft Office or Google Docs and launched an asynchronous coding tool, it’s about getting people hooked on the platform not the tech.

This will be a problem for all the AI providers to solve: when the models have roughly the same capabilities at competitive prices, what is your differentiator?

1. Claude Code is the one I miss the most when unavailable, but Gemini 2.5 Pro and OpenAI’s o3 are reasonable substitutes. ↩︎
2. Gemini 2.5 Pro, o3-pro, Sonnet/Opus 4 ↩︎
3. Claude can do this through MCP servers, but those take moderate technical know-how to setup. ChatGPT’s integrations just work. ↩︎
4. Google arguably has better voice models, but not a user-friendly interface. ↩︎