When you’re at the top, competitors will always find ways to challenge your dominance. OpenAI is clearly the leader, recently announcing 200 million people use Chat GPT every week, and that 92% of Fortune 500 companies are using its product and that API use has doubled since the release of GPT4o mini in July, according to a report from Axios.
Chat GPT is clearly in the lead as other providers have quickly caught up in terms of quality and features. So, how do you take down an established leader. That’s the question a lot of the competition is asking, and you can see different strategies at a play.
The Context Window Weakness
The easiest strategy is to target one of their weaknesses: their size. 200 million users, plus a popular API creates massive amounts of compute time and other companies can leverage their smaller compute costs to offer things that OpenAI can’t.
You can really see this clearly when you take a look at the anemic context windows ChatGPT offers. As you prompt AI Models, they store your prompts and their answers into a type of memory that keeps on being added to as your conversations continue.
An 8k context window on their free tier means it can process about 10 pages worth of text. That isn’t just the text you paste, but also the entirety of the conversation (the prompts and answers) you have with the model.
Here’s some common context window sizes for famous books. Even the paid tier can barely handle something like Doctor Jekyll and Mr. Hyde, and even the enterprise tier can’t do a book as long as Pride and Prejudice.
Meanwhile, the free tier from Claude offers a best in class 200k context window, and their models can handle up to a million or more tokens through the API. Google is not far behind offering a 32k context window in their free tier through Google Gemini. If you have Gemini Advanced that goes up to 1M. This is fine for a company with almost infinite compute like google who doesn’t have 200 million users per week hitting servers.
The Quality Attack
Another way in which these competitors are chipping away at OpenAI is the fact that they have all essentially caught up to Chat GPT in terms of quality. If you look at the LMSYS Chatbot Arena Leaderboard, you’ll see that although Open AI has managed to stay on top by releasing smaller improvements to their models, the competition has managed to get almost imperceptibly close to Chat GPT in terms of quality.
Heck, even Elon Musk’s Grok, a newcomer to the arena is ranking in the top 5. No matter OpenAI’s much awaited announcements of their new models…Strawberry, Q/Star, Orion or whatever they’re calling it. I can foresee these other companies catching up and surpassing ChatGPT.
Feature Weakness
This year, something I wasn’t expecting happened, companies started releasing features that were objectively better than ChatGPT’s. Case in point, Anthropic’s Claude released Artifacts, a feature that goes way beyond the Code Interpreter/Data Analysis tool in ChatGPT. It lets you chat with the AI and ask it to build code that you can live preview through a window. Combined with a newer 3.5 Sonnet model, it makes Anthropic a compelling alternative. Don’t get me wrong Claude is lacking things like Excel file uploads, image generation and others, but this is a visually compelling feature that sets is apart and above.
At the same time Google Gemini Advanced offers huge context windows and access to extensions like YouTube that lets you search through and chat about the platform’s videos. Plus it’s G button that lets you verify your results is a great addition lacking in just about every other tool.
The Image Edge
Recently Google added their Imagen 3 model, which gives you image generations (Chat GPT gives you an anemic 2 images per day on their free tier). Image generation and vision and essential to a complete AI experience. Clearly Midjourney is far ahead it’s competition here, launching their website with the best generation experience as well as features and quality way ahead of what others offer. Although DALL*E 3 is arguably better at prompt fidelity. You can’t argue with Midjourney’s creative chops.
The New Search Race
There’s an active race for a better search engine. ChatGPT has had a web access feature for quite a while, but it’s a struggle to get it to retrieve accurate, up to date content that’s relevant. In addition, the fact that all AIs have a hallucination problem means that you often can trust the returns.
The company that has built a search focused AI assistant that is clearly superior is Perplexity.ai. It’s so much better that a lot of people, tired of an ad filled Google results experience often use this as their search engine of choice. Google with it’s verify button and it’s access to the best search data in the market is easily better than what ChatGPT can do.
Conclusion
OpenAI has had all of the first to market advantages, but it’s gotten some serious competition with deep pockets. The user base might keep it on top for quite a while, and who knows if the new models will make a ton of difference. You can already see the competition chipping away at it’s dominance. I honestly don’t know if they’ll be able to hold onto the top much longer.
At least, competition is always good for the consumer, one of the results has been dramatically lower costs, context windows and features. I can’t way for the next wave.
